U.S. patent application number 14/853940 was filed with the patent office on 2016-08-18 for methods and system for linking geometry obtained from images.
This patent application is currently assigned to Bio-Tree Systems, Inc.. The applicant listed for this patent is Bio-Tree Systems, Inc.. Invention is credited to Raul A. Brauner, Kongbin Kang, Yanchun Wu.
Application Number | 20160239956 14/853940 |
Document ID | / |
Family ID | 51537574 |
Filed Date | 2016-08-18 |
United States Patent
Application |
20160239956 |
Kind Code |
A1 |
Kang; Kongbin ; et
al. |
August 18, 2016 |
METHODS AND SYSTEM FOR LINKING GEOMETRY OBTAINED FROM IMAGES
Abstract
Techniques for linking geometry extracted from one or more
medical images, the geometry including a plurality of geometric
objects each having parameter values including at least one value
for location and at least one value for direction/orientation, the
plurality of geometric objects comprising a target geometric object
and at least two candidate geometric objects, the techniques
include: (A) comparing parameter values of the target geometric
object with parameter values of the at least two candidate
geometric objects, (B) selecting one of the at least two candidate
geometric objects to link to the target geometric object based, at
least in part, on the comparison; and (C) linking the target
geometric object with the selected candidate geometric object.
Inventors: |
Kang; Kongbin; (Providence,
RI) ; Wu; Yanchun; (Sharon, MA) ; Brauner;
Raul A.; (Framingham, MA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Bio-Tree Systems, Inc. |
Framingham |
MA |
US |
|
|
Assignee: |
Bio-Tree Systems, Inc.
Framingham
MA
|
Family ID: |
51537574 |
Appl. No.: |
14/853940 |
Filed: |
September 14, 2015 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
PCT/US2014/028183 |
Mar 14, 2014 |
|
|
|
14853940 |
|
|
|
|
61791870 |
Mar 15, 2013 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
A61B 6/032 20130101;
G06T 7/70 20170101; G06T 17/005 20130101; G06K 9/6214 20130101;
G06T 7/181 20170101; G06T 2207/10081 20130101; G06T 7/11 20170101;
A61B 6/504 20130101; G06K 9/6202 20130101; G06T 7/12 20170101; G06T
2207/30101 20130101; A61B 6/5247 20130101; G06T 7/187 20170101;
G06K 9/6212 20130101; G06T 7/0012 20130101 |
International
Class: |
G06T 7/00 20060101
G06T007/00; G06T 17/00 20060101 G06T017/00; G06K 9/62 20060101
G06K009/62; A61B 6/00 20060101 A61B006/00; A61B 6/03 20060101
A61B006/03 |
Claims
1. A method for linking geometry extracted from one or more medical
images, the geometry including a plurality of geometric objects
each having parameter values including at least one value for
location and at least one value for direction/orientation, the
plurality of geometric objects comprising a target geometric object
and at least two candidate geometric objects, the method
comprising: (A) comparing parameter values of the target geometric
object with parameter values of the at least two candidate
geometric objects at least in part by: comparing at least one value
for location of the target geometric object to respective values
for location of the at least two candidate geometric objects, and
comparing at least one value for direction/orientation of the
target geometric object to respective values for
direction/orientation of the at least two candidate geometric
objects, (B) selecting one of the at least two candidate geometric
objects to link to the target geometric object based, at least in
part, on the comparison; and (C) linking the target geometric
object with the selected candidate geometric object.
2. The method of claim 1, wherein each of the plurality of
geometric objects further has at least one value for scale, and
wherein (A) further comprises: comparing at least one value for
scale of the target geometric object to respective values for scale
of the at least two candidate geometric objects.
3. The method of claim 2, wherein each of the plurality of
geometric objects further has at least one value for response of a
scale detection filter, and wherein (A) further comprises:
comparing at least one value for response of the scale detection
filter of the target geometric object to respective values for
response of the scale detection filter of the at least two
candidate geometric objects.
4. The method of claim 1, wherein the geometry represents a vessel
network and the target geometric object represents a cross-section
of a vessel structure in the vessel network, and wherein (A) is
performed by using a statistical model that provides a likelihood
that a candidate geometric object of the plurality of geometric
objects follows the target geometric object as a geometric
representation of another cross-section of the vessel structure
based, at least in part, on the at least one location value and the
at least direction/orientation value of the target object and at
least one location value and at least one direction orientation
value of the candidate geometric object.
5. The method of claim 4, wherein the statistical model provides
the likelihood that the candidate geometric object of the plurality
of geometric objects follows the target geometric object as a
geometric representation of another cross-section of the vessel
structure further based on at least one value for scale of the
target geometric object and at least one value for scale of the
candidate geometric object.
6. The method of claim 4, wherein the statistical model provides a
probability for parameters of a candidate geometric object
conditioned on parameters of the target geometric object.
7. The method of claim 1, wherein comparing the at least one value
for direction/orientation of the target geometric object to
respective values for direction/orientation of the at least two
candidate geometric objects is performed by using a super-Gaussian
probability model.
8. The method of claim 1, further comprising: calculating the at
least one value for direction/orientation of the target object
based, at least in part, on location information of voxels in at
least one segmented image.
9. The method of claim 8, wherein the at least one segmented image
includes at least one scale image.
10. The method of claim 8, wherein the calculating further
comprises computing displacement vectors between at least one voxel
location associated with the target geometric object and at least
one voxel location in a neighborhood associated with the target
geometric object.
11. The method of claim 10, further comprising performing principal
component analysis on a matrix formed from the computed
displacement vectors.
12. The method of claim 11, wherein the at least one value for
orientation is related to an eigenvector of the matrix.
13. An apparatus for linking geometry extracted from one or more
medical images, the geometry including a plurality of geometric
objects each having parameter values including at least one value
for location and at least one value for direction/orientation, the
plurality of geometric objects comprising a target geometric object
and at least two candidate geometric objects, the apparatus
comprising: at least one processor configured to perform: (A)
comparing parameter values of the target geometric object with
parameter values of the at least two candidate geometric objects at
least in part by: comparing at least one value for location of the
target geometric object to respective values for location of the at
least two candidate geometric objects, and comparing at least one
value for direction/orientation of the target geometric object to
respective values for direction/orientation of the at least two
candidate geometric objects, (B) selecting one of the at least two
candidate geometric objects to link to the target geometric object
based, at least in part, on the comparison; and (C) linking the
target geometric object with the selected candidate geometric
object.
14-15. (canceled)
16. The apparatus of claim 13, wherein the geometry represents a
vessel network and the target geometric object represents a
cross-section of a vessel structure in the vessel network, and
wherein the at least one processor is configured to perform (A) by
using a statistical model that provides a likelihood that a
candidate geometric object of the plurality of geometric objects
follows the target geometric object as a geometric representation
of another cross-section of the vessel structure based, at least in
part, on the at least one location value and the at least
direction/orientation value of the target object and at least one
location value and at least one direction orientation value of the
candidate geometric object.
17-19. (canceled)
20. The apparatus of claim 13, wherein the at least one processor
is configured to perform: calculating the at least one value for
direction/orientation of the target object based, at least in part,
on location information of voxels in at least one segmented
image.
21. (canceled)
22. The apparatus of claim 20, wherein the calculating further
comprises computing displacement vectors between at least one voxel
location associated with the target geometric object and at least
one voxel location in a neighborhood associated with the target
geometric object.
23-24. (canceled)
25. At least one non-transitory computer readable medium storing
instructions that, when executed by at least one processor, perform
a method of linking geometry extracted from one or more medical
images, the geometry including a plurality of geometric objects
each having parameter values including at least one value for
location and at least one value for direction/orientation, the
plurality of geometric objects comprising a target geometric object
and at least two candidate geometric objects, the method
comprising: (A) comparing parameter values of the target geometric
object with parameter values of the at least two candidate
geometric objects at least in part by: comparing at least one value
for location of the target geometric object to respective values
for location of the at least two candidate geometric objects, and
comparing at least one value for direction/orientation of the
target geometric object to respective values for
direction/orientation of the at least two candidate geometric
objects, (B) selecting one of the at least two candidate geometric
objects to link to the target geometric object based, at least in
part, on the comparison; and (C) linking the target geometric
object with the selected candidate geometric object.
26-27. (canceled)
28. The at least one non-transitory computer-readable medium of
claim 25, wherein the geometry represents a vessel network and the
target geometric object represents a cross-section of a vessel
structure in the vessel network, and wherein (A) is performed by
using a statistical model that provides a likelihood that a
candidate geometric object of the plurality of geometric objects
follows the target geometric object as a geometric representation
of another cross-section of the vessel structure based, at least in
part, on the at least one location value and the at least
direction/orientation value of the target object and at least one
location value and at least one direction orientation value of the
candidate geometric object.
29. (canceled)
29. (canceled)
30. (canceled)
31. The at least one non-transitory computer-readable medium of
claim 25, further comprising: calculating the at least one value
for direction/orientation of the target object based, at least in
part, on location information of voxels in at least one segmented
image.
32. (canceled)
33. The at least one non-transitory computer-readable medium of
claim 31, wherein the calculating further comprises computing
displacement vectors between at least one voxel location associated
with the target geometric object and at least one voxel location in
a neighborhood associated with the target geometric object.
34-35. (canceled)
36. The method of claim 1, further comprising evaluating at least
one quantitative or qualitative assessment of vascular morphology
based, at least in part, on linked geometry.
Description
RELATED APPLICATIONS
[0001] This application claims the benefit under 35 U.S.C.
.sctn.120 and is a continuation application of international
application number PCT/US2014/028183, filed Mar. 14, 2014, which
claims the benefit under 35 U.S.C. .sctn.119(e) of U.S. provisional
application No. 61/791,870, filed Mar. 15, 2013, each of which is
incorporated herein by reference in its entirety.
BACKGROUND
[0002] A wide range of imaging methods and devices are commonly
used to evaluate different anatomical and physiological conditions
in a variety of medical and research environments. Tools have been
developed to image body structures based on different physical
properties. For example, X-rays, CT scans, MRIs, PET scans, IR
analyses and other technologies have been developed to obtain
images of various body structures. These tools are routinely used
for diagnostic, therapeutic, and research applications.
Combinations of two or more different imaging techniques are
sometimes used to provide complementary information about a
patient.
[0003] In conventional medical imaging, a human operator, such as a
physician or diagnostician, may visually inspect one or more images
to make an assessment, such as detection of a tumor or other
pathology or to otherwise characterize the internal structures of a
patient. However, this process may be difficult and time consuming.
For example, it may be difficult to assess 3D biological structure
by attempting to follow 2D structure through a series of stacked 2D
images. In particular, it may be perceptually difficult and time
consuming to understand how 2D structure is related to 3D structure
as it appears, changes in size and shape, and/or disappears in
successive 2D image slices. A physician may have to mentally
arrange hundreds or more 2D slices into a 3D picture of the
anatomy. To further frustrate this process, when anatomical
structure of interest is small, the structure may be difficult to
discern or it may be difficult to understand how numerous
structures relate to a biological whole.
[0004] Furthermore, in addition to the time consuming nature of
manual inspection, human visual interpretation of images has
further shortcomings. While the human visual cortex processes image
information to obtain qualitative information about structure in
the image, it does not compute quantitative geometry from the
image. However, the quantitative geometry of the structure
represented in one or more images may contain valuable information
about the structure that can be used to diagnose disease, assess
the efficacy of treatment and/or perform other analyses of the
structure. Such quantitative information about the structure is
beyond the capability of conventional human visual image
understanding alone.
[0005] Image processing techniques have been developed to automate
or partially automate the task of understanding and partitioning
the structure in an image and are employed in computer aided
diagnosis (CAD) to assist a physician in identifying and locating
structure of interest in a 2D or 3D image. CAD techniques often
involve segmenting the image into groups of related pixels and
identifying the various groups of pixels, for example, as those
comprising a tumor or a vessel or some other structure of interest.
However, conventional segmentation may produce unsatisfactory or
incomplete results, particularly when the structure being detected
appears in the image at arbitrary locations, sizes and
orientations. As a result, the limited geometry that may be
extracted from conventional image processing may be unsuitable for
use in further analysis based on the extracted geometry.
SUMMARY
[0006] The inventors have developed methods and apparatus for
extracting geometry from images, scan data, and/or representations
of tubular body structures (e.g., blood vessels or other body
vessels). Aspects of the technology described herein relate to
obtaining vessel geometry, determining one or more structural
features from the vessel geometry, and/or analyzing the one or more
structural features for medical diagnostic, prognostic, and/or
research applications.
[0007] The inventors have developed methods and apparatus for
extracting geometry from images, scan data, and/or representations
of tubular body structures (e.g., blood vessels or other body
vessels). Aspects of the technology described herein are useful for
obtaining a geometrical representation of a vascular tree that
contains data relating to three-dimensional location, orientation
and/or size at any point in the vascular tree of a subject. In some
embodiments, a vascular tree may be represented by a series of
disks or poker chips (e.g., circular or elliptical disks) that are
linked together to form a three-dimensional structure containing
information relating to the local size, shape, branching, and other
structural features at any point in the vascular tree.
[0008] It should be appreciated that the entire vascular tree of a
subject may be represented by a network of linked poker chips
(e.g., circular or elliptical disks). However, in many embodiments,
only a subset or a portion of a vascular tree may be represented or
analyzed. In some embodiments, a portion of a vascular tree can be
represented by a single disc or poker chip that contains
information relating to the location of the center of the vessel,
vessel size (diameter), and/or orientation (e.g., the direction of
the centerline of the vessel). In some embodiments, a portion of a
vascular tree may be represented by a dataset that describes one or
more poker chips along with information relating to the linkage
between the poker chips within a region of interest of the vascular
tree.
[0009] Some embodiments are directed to an apparatus for linking
geometry extracted from one or more medical images, the geometry
including a plurality of geometric objects each having parameter
values including at least one value for location and at least one
value for direction/orientation, the plurality of geometric objects
comprising a target geometric object and at least two candidate
geometric objects. The apparatus comprises at least one processor
configured to perform: (A) comparing parameter values of the target
geometric object with parameter values of the at least two
candidate geometric objects at least in part by: comparing at least
one value for location of the target geometric object to respective
values for location of the at least two candidate geometric
objects, and comparing at least one value for direction/orientation
of the target geometric object to respective values for
direction/orientation of the at least two candidate geometric
objects, (B) selecting one of the at least two candidate geometric
objects to link to the target geometric object based, at least in
part, on the comparison; and (C) linking the target geometric
object with the selected candidate geometric object.
[0010] Some embodiments are directed to at least one non-transitory
computer readable medium storing instructions that, when executed
by at least one processor, perform a method of linking geometry
extracted from one or more medical images, the geometry including a
plurality of geometric objects each having parameter values
including at least one value for location and at least one value
for direction/orientation, the plurality of geometric objects
comprising a target geometric object and at least two candidate
geometric objects, the method comprising: (A) comparing parameter
values of the target geometric object with parameter values of the
at least two candidate geometric objects at least in part by:
comparing at least one value for location of the target geometric
object to respective values for location of the at least two
candidate geometric objects, and comparing at least one value for
direction/orientation of the target geometric object to respective
values for direction/orientation of the at least two candidate
geometric objects, (B) selecting one of the at least two candidate
geometric objects to link to the target geometric object based, at
least in part, on the comparison; and (C) linking the target
geometric object with the selected candidate geometric object.
[0011] Some embodiments are directed to a method of linking
geometry extracted from one or more medical images, the geometry
including a plurality of geometric objects each having parameter
values including at least one value for location and at least one
value for direction/orientation, the plurality of geometric objects
comprising a target geometric object and at least two candidate
geometric objects, the method comprising: (A) comparing parameter
values of the target geometric object with parameter values of the
at least two candidate geometric objects at least in part by:
comparing at least one value for location of the target geometric
object to respective values for location of the at least two
candidate geometric objects, and comparing at least one value for
direction/orientation of the target geometric object to respective
values for direction/orientation of the at least two candidate
geometric objects, (B) selecting one of the at least two candidate
geometric objects to link to the target geometric object based, at
least in part, on the comparison; and (C) linking the target
geometric object with the selected candidate geometric object.
[0012] In some embodiments, each of the plurality of geometric
objects further has at least one value for scale, and (A) further
comprises comparing at least one value for scale of the target
geometric object to respective values for scale of the at least two
candidate geometric objects.
[0013] In some embodiments, including any of the preceding
embodiments, each of the plurality of geometric objects further has
at least one value for response of a scale detection filter, and
wherein (A) further comprises: comparing at least one value for
response of the scale detection filter of the target geometric
object to respective values for response of the scale detection
filter of the at least two candidate geometric objects.
[0014] In some embodiments, including any of the preceding
embodiments, the geometry represents a vessel network and the
target geometric object represents a cross-section of a vessel
structure in the vessel network, and wherein (A) is performed by
using a statistical model that provides a likelihood that a
candidate geometric object of the plurality of geometric objects
follows the target geometric object as a geometric representation
of another cross-section of the vessel structure based, at least in
part, on the at least one location value and the at least
direction/orientation value of the target object and at least one
location value and at least one direction orientation value of the
candidate geometric object.
[0015] In some embodiments, including any of the preceding
embodiments, the statistical model provides the likelihood that the
candidate geometric object of the plurality of geometric objects
follows the target geometric object as a geometric representation
of another cross-section of the vessel structure further based on
at least one value for scale of the target geometric object and at
least one value for scale of the candidate geometric object.
[0016] In some embodiments, including any of the preceding
embodiments, the statistical model provides a probability for
parameters of a candidate geometric object conditioned on
parameters of the target geometric object.
[0017] In some embodiments, including any of the preceding
embodiments, comparing the at least one value for
direction/orientation of the target geometric object to respective
values for direction/orientation of the at least two candidate
geometric objects is performed by using a super-Gaussian
probability model.
[0018] In some embodiments, including any of the preceding
embodiments, the method further comprises calculating the at least
one value for direction/orientation of the target object based, at
least in part, on location information of voxels in at least one
segmented image.
[0019] In some embodiments, including any of the preceding
embodiments, the at least one segmented image includes at least one
scale image.
[0020] In some embodiments, including any of the preceding
embodiments, the calculating further comprises computing
displacement vectors between at least one voxel location associated
with the target geometric object and at least one voxel location in
a neighborhood associated with the target geometric object.
[0021] In some embodiments, including any of the preceding
embodiments, the method further comprises performing principal
component analysis on a matrix formed from the computed
displacement vectors.
[0022] In some embodiments, including any of the preceding
embodiments, the at least one value for orientation is related to
an eigenvector of the matrix.
[0023] Some embodiments include a method of computing
direction/orientation of a geometric object extracted from CT
information using at least one segmented image computed from the CT
information, the method comprising determining at least one
displacement vector from a voxel location associated with the
geometric object and at least one other voxel location in a
neighborhood associated with the geometric object, and determining
a direction/orientation of the geometric object based, at least in
part, on the at least one displacement vector. According to some
embodiments, the at least one segmented image includes at least one
scale image.
[0024] Some embodiments of methods for computing
direction/orientation include performing principal component
analysis on a matrix formed from the at least one displacement
vector. According to some embodiments, the direction/orientation is
related to an eigenvector of the matrix.
[0025] Some embodiments includes at least one computer readable
medium storing instructions that, when executed by at least one
processor, perform a method of computing direction/orientation of a
geometric object extracted from CT information using at least one
segmented image computed from the CT information, the method
comprising determining at least one displacement vector from a
voxel location associated with the geometric object and at least
one other voxel location in a neighborhood associated with the
geometric object, and determining a direction/orientation of the
geometric object based, at least in part, on the at least one
displacement vector.
[0026] Some embodiments include an apparatus for computing
direction/orientation of a geometric object extracted from CT
information using at least one segmented image computed from the CT
information, the apparatus comprising at least one processor
configured to determine at least one displacement vector from a
voxel location associated with the geometric object and at least
one other voxel location in a neighborhood associated with the
geometric object, and determine a direction/orientation of the
geometric object based, at least in part, on the at least one
displacement vector.
[0027] Some embodiments include a method of determining a branch
point candidate corresponding to a location where a vessel
structure branches, the branch point determined from geometry
extracted from CT information that comprises a plurality of
geometric objects including a first geometric object, the method
comprising determining at least one displacement vector from a
voxel location associated with the first geometric object and at
least one other voxel location in a neighborhood associated with
the first geometric object, and determining at least one value
indicative of an asymmetry at the first geometric object based, at
least in part, on the at least one displacement vector.
[0028] According to some embodiments, the branch point is
determined using at least one segmented image, and according to
some embodiments, the at least one segmented image includes at
least one scale image. Some embodiments of methods of determining a
branch point candidate include performing principal component
analysis on a matrix derived from the at least one displacement
vector. According to some embodiments, the at least one value
indicative of an asymmetry is related to an eigenvalue of one of
the eigenvectors of the matrix.
[0029] Some embodiments include at least one computer readable
medium storing instructions that, when executed by at least one
processor, performs a method of determining a branch point
candidate corresponding to a location where a vessel structure
branches, the branch point determined from geometry extracted from
CT information that comprises a plurality of geometric objects
including a first geometric object, the method comprising
determining at least one displacement vector from a voxel location
associated with the first geometric object and at least one other
voxel location in a neighborhood associated with the first
geometric object, and determining at least one value indicative of
an asymmetry at the first geometric object based, at least in part,
on the at least one displacement vector.
[0030] Some embodiments include an apparatus for determining a
branch point candidate corresponding to a location where a vessel
structure branches, the branch point determined from geometry
extracted from CT information that comprises a plurality of
geometric objects including a first geometric object, the method
comprising at least one processor configured to determine at least
one displacement vector from a voxel location associated with the
first geometric object and at least one other voxel location in a
neighborhood associated with the first geometric object, and
determine at least one value indicative of an asymmetry at the
first geometric object based, at least in part, on the at least one
displacement vector.
[0031] Some embodiments include methods for detecting and resolving
loops in vessel so that the linked vessel structure (e.g., a
directed or non-directed graph) accurately represents loops in the
vessel structure (e.g., the graph structure may be cyclic).
According to some embodiments, loops are detected in part by
labeling Poker Chips.TM. as visited and/or linked such that when a
Poker Chip.TM. that is labeled as visited and/or linked is
identified as a link candidate for more than a single link
structure, the Poker Chip.TM. can be evaluated from both directions
to assess whether the vessel structure forms a loop.
[0032] Some embodiments include accelerating linking by dividing a
geometric representation and associated image data (e.g.,
intensity, segmented, scale image(s), etc.) into smaller regions
and processing them in parallel. The inventors have developed
techniques for stitching the linked structures from the smaller
regions together to form a larger linked structure representing the
vessel network. According to some embodiments, location and
direction of Poker Chips.TM. in a glue region at the juncture of
adjacent regions are evaluated to determine how sub-structures
should be stitched or glued together to form a larger linked
structure.
[0033] According to aspects of the technology described herein, a
poker chip representation of a vasculature may be mined for
physiological, biological, and/or medical purposes. In some
embodiments, geometrical information associated with a single poker
chip may be mined. In some embodiments, geometrical information
associated with a plurality of poker chips, optionally including
local linkage information may be mined.
Accordingly, aspects of the technology described herein relate to
obtaining vessel geometry, determining one or more structural
features from the vessel geometry, and/or analyzing the one or more
structural features for medical diagnostic, prognostic, and/or
research applications.
[0034] Aspects of the technology described herein provide methods
for analyzing structures such as blood vessels and evaluating their
association with disease, responsiveness to therapeutic treatments,
and/or other conditions. Aspects of the technology described herein
provide quantitative and analytical methods for evaluating and/or
comparing the vessels in different regions of the same body (e.g.,
a human body) or within ex vivo tissues or between different bodies
(e.g., the same regions in different bodies) or different ex vivo
tissues. Aspects of the technology described herein can be useful
in assisting and/or automating the analysis of vascular patterns
and their association with disease diagnosis, prognosis, response
to therapy, toxicity evaluation, etc., or any combination thereof.
Aspects of the technology described herein can be used in
connection with vessel structural information that is obtained from
vessel images (e.g., blood vessel images), scan data, vessel
representations (e.g., a reconstructed vasculature, a
representation that can be viewed as being similar in some ways to
a stack of poker chips with varying diameters and is that is
referred to herein as a Poker Chip representation, or any other
useful representation, or any combination thereof).
[0035] Methods are provided for analyzing vessel structural
features, and blood vessel structural features in particular. In
some embodiments, a distribution of vessel parameters (e.g.,
structural features or morphological parameters) within a region of
interest may be generated and evaluated. In some embodiments, the
vessel parameters may relate to the size, shape, or number of
vessels with a region of interest. A distribution may be generated
based on quantitative measurements related to one or more
parameters. In some embodiments, a distribution of blood vessels
may be a population distribution of blood vessels as a function of
quantitative measures of one or more parameters. For example, a
distribution may represent the number of blood vessels (or the
percentage of the blood vessel population) as a function of their
diameter, branching frequency, distance between branches, degree of
tortuousity, curvature, or any other quantitative structural
feature or morphological parameter, e.g., as described herein, or
any combination of two or more thereof. In some embodiments, a
distribution may be divided into groups or bins representing
different value ranges of the quantitative measurements (e.g.,
ranges of vessel diameters such as 0-30 microns, 30-60 microns,
60-90 microns, 90-120 microns, 120-150 microns, 150-180 microns,
etc., or any combination thereof). It should be appreciated that a
distribution may be represented in any suitable form, for example
graphically (e.g., a graph or histogram), in the form of a table,
as a database, in a computer-readable or computer storage medium,
etc., or any combination thereof.
BRIEF DESCRIPTION OF THE DRAWINGS
[0036] FIG. 1 illustrates a flow chart of extracting geometry from
an image, in accordance with some embodiments of the technology
described herein;
[0037] FIG. 2 illustrates a geometrical representation of vessel
structure, referred to as the Poker Chip.TM. representation, in
accordance with some embodiments of the technology described
herein;
[0038] FIG. 3A illustrates a cylindrical segment used to model
vessel structure, in accordance with some embodiments of the
technology described herein;
[0039] FIG. 3B illustrates a grey scale representation of a
characteristic function of a model used to detect vessel
structures, in accordance with some embodiments of the technology
described herein;
[0040] FIG. 3C illustrates a plot of the intensity values along the
x-axis at the center of the grey scale Gaussian distribution in
FIG. 3B;
[0041] FIG. 3D illustrates a plot of the intensity values along the
x-axis of another model of vessel intensity profile;
[0042] FIG. 4 illustrates schematically a cylindrical vessel
segment intensity distribution illustrating a ridge or centerline
feature, in accordance with some embodiments of the technology
described herein;
[0043] FIG. 5 illustrates an embodiment of a mixture of truncated
Gaussian fit to 3D reconstruction intensity data, wherein the
vertical axis is in log scale and low part of the horizontal axis
is shown;
[0044] FIG. 6 illustrates an embodiment of a theoretical profile of
a centerline filter response using scale detection, in accordance
with some embodiments of the technology described herein;
[0045] FIG. 7 illustrates an embodiment of a detected scale versus
the choice of threshold .alpha.;
[0046] FIG. 9 illustrates an embodiment of how R(X, r) behaviors on
real images--(a) a slice of 3D images is shown and blue point is
the point X where we apply rank-based scale filter--(b) the
rank-based scale filter's response with different radius is
shown--although the intensities have large variation inside vessel,
the rank-based scale filter behavior smoothly and have a rapidly
decay while cross the boundary of the vessel;
[0047] FIG. 10A illustrates a centerline filter, in accordance with
some embodiments of the technology described herein;
[0048] FIG. 10B illustrates a profile of the centerline filter
illustrated in FIG. 9A along the line x-x', in accordance with some
embodiments of the technology described herein;
[0049] FIG. 10C illustrates another profile of the centerline
filter illustrated in FIG. 9A along the line x-x', in accordance
with some embodiments of the technology described herein;
[0050] FIG. 11 illustrates centerline filtering on a 3D volume data
set, in accordance with some embodiments of the technology
described herein;
[0051] FIG. 12 illustrates net volume of the center line filter
versus different scales;
[0052] FIG. 13 illustrates a geometrical representation of
vasculature obtained from a 3D volumetric image, in accordance with
some embodiments of the technology described herein;
[0053] FIG. 14 illustrates blood vessel size distribution in an
example of casts of a xenograft tumor model after treatment with
Avastin.RTM. (an anti-angiogenic agent available from Genentech,
South San Francisco, Calif.), in accordance with some embodiments
of the technology described herein;
[0054] FIG. 15 illustrates the vessel population ratio between
small and middle size vessels in an example of casts of a xenograft
tumor model after treatment with Avastin.RTM., in accordance with
some embodiments of the technology described herein;
[0055] FIG. 16 illustrates the vessel population ratio between
large and middle size vessels in an example of casts of a xenograft
tumor model after treatment with Avastin.RTM., in accordance with
some embodiments of the technology described herein;
[0056] FIG. 17 illustrates the vessel population distribution in an
example of casts of a tumor model after treatment with
Avastin.RTM., in accordance with some embodiments of the technology
described herein;
[0057] FIG. 18 illustrates the vessel population ratio between
small and middle size vessels in an example of casts of a tumor
model after treatment with Avastin.RTM., in accordance with some
embodiments of the technology described herein;
[0058] FIG. 19 illustrates the vessel population ratio between
large and middle size vessels in an example of casts of a tumor
model after treatment with Avastin.RTM., in accordance with some
embodiments of the technology described herein;
[0059] FIG. 20 is a flowchart of an illustrative process for
generating a linked representation of a vessel network, in
accordance with some embodiments of the technology described
herein;
[0060] FIG. 21 is a flowchart of an illustrative process for
linking geometric objects that represent cross-sections of a vessel
in a vessel network, in accordance with some embodiments of the
technology described herein;
[0061] FIG. 22 is a flowchart of an illustrative process for
detecting branching point locations, in accordance with some
embodiments of the technology described herein;
[0062] FIG. 23 is a flowchart of an illustrative process for
calculating branching scores for locations on a centerline of a
vessel, in accordance with some embodiments of the technology
described herein;
[0063] FIGS. 24A, 24B, and 24C illustrate calculation of a
branching score, in accordance with some embodiments of the
technology described herein;
[0064] FIG. 25 is a flowchart of an illustrative process for
analyzing characteristics of branch point candidates, in accordance
with embodiments of the technology described herein;
[0065] FIGS. 26A, 26B, and 26C illustrate Y, V, and T structures of
vessel branch points, respectively, in accordance with some
embodiments of the technology described herein;
[0066] FIG. 27 illustrates a two-dimensional scale image, in
accordance with some embodiments of the technology described
herein; and
[0067] FIG. 28 is a block diagram of an illustrative computer
system that may be used in implementing some embodiments.
DETAILED DESCRIPTION
[0068] As discussed above, analyzing vessel structures (e.g., blood
vessel structures) and identifying structural profiles that are
characteristic of one or more physiological conditions or responses
(e.g., positive responses to pharmaceutical compounds) may be of
interest in many areas of diagnostics, therapeutics and/or
treatment. However, the amount of information that can be directly
obtained or ascertained from image data (e.g., x-ray, CT, MRI,
etc.) may be prohibitively limited in this respect. Accordingly,
the inventors have recognized the benefit of developing methods of
extracting geometry from images to facilitate the above described
analysis.
[0069] To extract geometrical properties of vessel structures in
one or more images, the vessels must first be detected in the image
and represented in a meaningful fashion. Various methods have been
proposed for detecting one or more features of a blood vessel using
a filter adapted to respond to the one or more features. For
example, filters have been designed to respond to the intensity
profile of a vessel to locate voxels that exhibit this intensity
profile. However, conventional filtering techniques may be
unsatisfactory at accurately and robustly detecting vessel
structures in one or more images. Filtering techniques typically
require some additional preprocessing to obtain information about
the image to improve the filtering process. For example, the scale
of the structure at a particular location in the image may be
obtained to determine what size filter should be used at that
location. That is, not only should the filter match the feature
being detected, in order to respond correctly, the filter should
also match the scale of the feature. Moreover, because the
orientation of the feature being detected is not known a priori,
filtering techniques often include some preprocessing to determine
the orientation of the feature at a particular location so that the
filter can be applied to the image in general alignment with the
feature.
[0070] Conventionally, scale detection and orientation detection
are performed simultaneously. The inventors have appreciated that
simultaneous scale and orientation detection may result in
sub-optimal detection of either scale, orientation or both. As a
result, subsequent filtering to detect one or more features applied
using sub-optimal scale and orientation parameters may be
substantially degraded. The inventors have developed a method for
detecting vessel features that includes a scale detection operation
and an orientation detection operation that are performed
separately. In some embodiments, scale detection is performed prior
to orientation detection, and orientation detection is performed
using the scale determined by the scale detection. The scale and
orientation values determined from the separate scale and
orientation detection operations may then be used to apply the
feature detection filter, for example, a centerline filter adapted
to respond to the centerline voxels of blood vessels.
[0071] According to some embodiments, scale detection employs an
orientation independent scale detector such that scale detection
may be performed independent of orientation detection. According to
some embodiments, an orientation independent scale filter is used
having a filter kernel that is symmetric with respect to
orientation such that the filter does not rely on orientation for
accurate scale detection. According to some embodiments, the
orientation independent scale filter includes a filter size defined
by a radius. At each of a plurality of selected voxels in an image,
the orientation independent scale filter is applied at increasing
radii until the filter response fails to meet a predetermined
criteria. The largest radius at which the filter response meets the
predetermined criteria is used to represent the scale. According to
some embodiments, the diameter of vessel structures in the images
is determined based on this largest radius. That is, according to
some embodiments, at least some geometry of vessel structures may
be determined by the scale detection operation.
[0072] The inventors have appreciated that performing scale
detection, orientation detection and centerline detection provides,
at each detected centerline voxel, the location, the direction of
the centerline and the radius of the vessel. This geometry can be
used to analyze vascular structure and these geometrical parameters
have been used to develop a mathematical representation of the
detected vessel structure. In some embodiments, each centerline
location may be represented as a circular or elliptical disk having
a center at the centerline location, a radius corresponding to the
associated scale, and a normal vector to the disk (e.g., circular
disk) corresponding to the direction of the centerline as
determined during orientation detection. This representation
resembles a poker chip and is referred to herein as the Poker
Chip.TM. representation, as described in further detail below.
[0073] While the Poker Chip.TM. representation provides much useful
information about the geometry of the vessel, without further
processing, there is no notion of adjacency or vessel membership,
which may be useful information in performing analysis on the
vasculature. Accordingly, in some embodiments, each of the detected
centerline voxels (e.g., center locations of a poker chip) are
linked together to capture adjacency information as well as vessel
membership. In some embodiments, the centerline voxels are linked
according to a criteria that includes one or any combination of
minimizing a distance, a direction change, a radius change, and/or
a filter response change from a centerline voxel to an adjacent
centerline voxel. That is, when selecting between a number of
candidate centerline voxels to link to a target centerline voxel,
the centerline voxel candidate that creates the smallest change in
one or more of the above parameters may be preferred over candidate
centerline voxels having larger changes. The linked centerline
voxels can then be used to compute various structural
characteristics of the vasculature formed by the detected vessels
as represented by the stacked and linked poker chips.
[0074] To generate more comprehensive linked structures, points at
which vessels branch may be detected so that vessel centerlines
from branched vessels can be appropriately linked together. The
inventors have appreciated that branch points may often exhibit an
asymmetric property associated with the detected centerline points.
In view of this insight, the inventors have developed techniques to
detect at least one indication of asymmetry to identify branch
point candidates. According to some embodiments, detecting the at
least one indication of asymmetry comprises performing principal
component analysis on a neighborhood of respective target
centerline voxels detected from an image (e.g., a 3D image) of
vasculature. According to some embodiments, the principal
directions of variation and/or their respective significance may be
evaluated to assess the symmetry/asymmetry of the neighborhood of a
target centerline voxel to determine the likelihood that a branch
point is present. For example, the eigenvectors and/or associated
eigenvalues computed from a matrix formed from the neighborhood of
a centerline voxel may be evaluated to detect at least one
indication of asymmetry at a location associated with the
centerline voxel. However, other measures of asymmetry may be
computed in other ways to identify branch point candidates, as the
aspects are not limited in this respect.
[0075] Following below are more detailed descriptions of various
concepts related to, and embodiments of, methods and apparatus
according to the present invention. It should be appreciated that
various aspects of the invention described herein may be
implemented in any of numerous ways. Examples of specific
implementations are provided herein for illustrative purposes only.
In addition, the various aspects of the invention described in the
embodiments below may be used alone or in any combination, and are
not limited to the combinations explicitly described herein.
[0076] FIG. 1 illustrates a method of extracting vessel geometry
from one or more images of vasculature, in accordance with some
embodiments of the technology described herein. Act 110 includes
obtaining image information of at least a portion of a vasculature
structure. For example, the image information may be a
two-dimensional (2D), three-dimensional (3D) or other dimensional
image obtained from scanning an object using x-ray CT, MRI, PET,
SPECT, etc. The scanned object may be a live specimen such as a
human or other animal (i.e., an in-vivo scan), or obtained from a
cast of a specimen's vasculature.
[0077] The method of FIG. 1 may be performed on any image of any
dimension independent of how the image was obtained, as the aspects
of the invention are not limited in this respect. In 2D images,
each 2D location having an associated intensity is conventionally
referred to as a pixel. In 3D images, each volume location having
an associated intensity is conventionally referred to as a voxel.
The term voxel is used herein to refer to both 2D and 3D image
locations to eliminate the need to specify the dimensionality of
the images, as the methods described herein are generic to
dimensionality.
[0078] Many techniques for extracting information from images use
various filtering techniques. For example, filters are often
designed such that when applied to a portion of an image (e.g.,
convolved with a portion of the image) the filter response is
relatively large when the filter is applied to an image portion
having a feature or characteristic indicative of structure being
detected in the image, and relatively small otherwise. The filter
detection described below in connection with act 140 is one example
of matched filtering. However, other filtering techniques may be
used, as the aspects of the technology described herein are not
limited in this respect.
[0079] When the feature or structure being detected appears in an
image at different sizes or scales, the size of the filter kernel
should be adjusted to the appropriate scale in order for the filter
response to accurately indicate the presence of the desired
feature. For example, in an image containing biological
vasculature, and in particular, tumor vasculature, the constituent
vessels will typically vary greatly in diameter. Accordingly, a
filter designed to detect relatively large vessels will not respond
accordingly to small vessels, even when applied on the correct
location. However, it is not known a priori where large and small
vessels are located. Accordingly, successful detection may require
determining the scale of the structure in the image prior to
applying the filter. This technique is herein referred to as "scale
detection." Scale detection may be performed on predetermined
portions of an image, or may be determined on a voxel by voxel
basis, as described in further detail below.
[0080] In addition to detecting the appropriate scale, it may be
beneficial to detect the orientation in which the filter should be
applied. In particular, the feature(s) being detected may appear in
the image at arbitrary orientations. For example, in the case of
vasculature, the vessel properties being detected may be oriented
in any arbitrary direction. Accordingly, even if a filter at the
appropriate scale is applied at an image region corresponding to
the feature being detected, the filter response may be relatively
low if it is not oriented in general alignment with the direction
of the feature for which the filter was designed to detect.
Accordingly, determining the orientation of the features or
properties being detected may benefit filter detection techniques.
This technique is herein referred to as "orientation
detection."
[0081] Conventional filtering techniques combine scale and
orientation detection in a single operation. That is, the
combination of possible scales and orientations are tested
simultaneously and the scale and orientation are selected when the
response is maximum. However, the inventors have appreciated that
maximum responses may not correspond to optimal scale and optimal
orientation simultaneously. Because the response is a combination
of scale and orientation, one or both may be sub-optimal while
together providing a strong response. The inventors have developed
a scale detection operation that is orientation independent. As a
result, the operations of scale detection and orientation detection
may be separated into two separate operations. In addition, the
detected scale may then be used to improve subsequent orientation
detection processes.
[0082] In act 120, scale detection is performed independently of
orientation detection. In some embodiments, scale detection 120 is
performed using a filter that is independent of orientation. Scale
detection 120 may provide the scale in the image at different
regions in the image. In some embodiments, scale detection 120
determines scale at each voxel in the image. Alternatively, a
preprocessing operation may be performed to roughly determine which
voxels in the image correspond to subject matter of interest (e.g.,
vessels) and which voxels correspond to background. Scale detection
may then be performed only on pixels determined to correspond to
subject matter of interest, thus reducing the amount of
computations. The result of scale detection is a scale associated
with each location at which the filter was applied (e.g., a scale
at each selected voxel in the image). An orientation independent
scale detection algorithm according to some embodiments is
described in further detail below.
[0083] In act 130, orientation detection may be performed. To
assist in more accurate orientation detection, the scale at the
selected regions of the image determined during scale detection 120
may be provided to the orientation detection operation. As
discussed above, determining the orientation of subject matter of
interest in one or more images may be important for accurate filter
detection of the subject matter of interest (e.g., structure,
feature, property or characteristic). For example, in embodiments
where the subject matter of interest is vasculature, it may be
important to detect the direction of the center or longitudinal
axis of the vessels before applying a filter that detects the
centerline of the vessel. In some embodiments, the scale determined
from scale detection 120 may be used to improve orientation
detection accuracy. The result of orientation detection is an
orientation or direction at each selected voxel indicating the
direction of the centerline at the respective location. An
orientation detection algorithm according to some embodiments is
described in further detail below.
[0084] In act 140, filter detection may be performed. In filter
detection 140, a filter designed to respond to the subject matter
of interest in the image may be applied. In some embodiments, the
filter is applied at the scale and/or orientation determined from
scale detection and/or orientation detection, respectively. The
magnitude of the filter response at selected locations in the image
indicates the likelihood that the location includes the subject
matter of interest. In some embodiments, the subject matter of
interest is vasculature and the filter is designed to respond to
the center of a vessel. That is, the filter may be designed to
respond to the intensity profile across a vessel and thus respond
most strongly when centered on a centerline voxel in the direction
of the intensity profile. Because the scale and direction of the
subject matter of interest has been determined at selected
locations in the image, filter detection may appropriately accurate
in detecting the subject matter of interest. Several methods of
centerline filtering are discussed in detail below, in accordance
with some embodiments of the technology described herein.
[0085] In act 150, non-maximal suppression may be performed on the
output of the filter detection operation performed in act 140. As
discussed above, the result of a filtering operation (e.g.,
centerline filtering) generally includes the filter response at
each voxel at which the filter was applied. The magnitude of the
response is typically proportional to the likelihood that the
feature being detected is present at the corresponding voxel
location. However, it should be appreciated that many voxel
locations will have associated non-zero filter responses. In
addition, some voxel locations will have associated local maximum
filter responses even though the true location of the feature is
elsewhere. However, accurate detection may require discriminating
between local maximum and the true maximum location, which
corresponds to the most likely location of the structure being
detected. Non-maximal suppression 150 attempts to eliminate or
suppress all but the true maximum filter responses to accurately
detect the subject matter of interest. A detailed description of
non-maximum suppression in the context of centerline filtering for
vessel detection is described below.
[0086] In act 160, linking may be performed. Linking may include
various operations that associate voxel locations with each other
to form related structures so that geometric properties may be
obtained from the linked voxels. For example, in the context of
vessel detection, the voxel locations that were determined as
centerline voxels after centerline detection and non-maximum
suppression may be linked together to form the associated
centerline of vessels. That is, analysis may be performed to link
together centerline voxels that are likely to have arisen from the
same vessel structure. In such a way, the geometry of the vessels
may be obtained (e.g., geometry 15). Methods for linking voxels in
the context of vessel detection are described in further detail
below.
[0087] As discussed above, some embodiments are directed to
detecting vasculature and extracting the geometry of the
vasculature to facilitate various analysis such as diagnosis,
therapeutics, drug efficacy, etc. The inventors have developed
methods for extracting geometrical information from 3D volumetric
images using a match filter based system to segment a vessel
network and extract a mathematical (geometry) vessel
representation. Some embodiments of a vessel representation are
referred to herein as the Poker Chip.TM. representation due to the
similarity to a stack of poker chips. The Poker Chip.TM.
representation treats a vessel as an aggregation of infinitesimal
cylinder cross-sections with continuously varying diameters. While
in theory the "thickness" of each poker chip is infinitesimal, in
practice the thickness of each poker chip may be related to the
resolution of the image(s) from which the geometry was extracted.
Thus, each poker chip may have associated geometry including, for
example, center location, radius and orientation, as discussed in
further detail below.
[0088] FIG. 2 illustrates a schematic of the Poker Chip.TM.
representation. According to some embodiments, each poker chip 210
is defined by a center location, a radius and an orientation. The
center location c.sub.i represents the center of the vessel, for
example, determined by centerline filtering, as discussed in
further detail below. The radius r represents the radius of the
vessel at location c.sub.i and the orientation is the angle of the
normal of the poker chip at location c.sub.i, and represents the
tangent of the centerline of the vessel at location c.sub.i. It
should be appreciated that the Poker Chip.TM. representation may
include additional parameters, as the aspects of the technology
described herein are not limited in this respect.
[0089] The inventors have appreciated that the above Poker Chip.TM.
representation may be used to determine characteristics of the
vasculature that may help in diagnosing disease, providing
information on appropriate treatment, and/or assessing the
effectiveness of treatment. For example, since the orientation is
known at each location, higher level information such as curvature
and tortuosity may be computed, as well as vessel density and
distribution measures, as discussed in further detail below.
Additionally, since vessel diameter may be determined, vessel size
and the change in vessel sizes may be computed as well. Various
analyses that can be performed using the Poker Chip.TM.
representation are discussed in further detail below.
[0090] To compute some of the higher order information, it may be
beneficial to also include in the Poker Chip.TM. representation
information about neighboring poker chips. For example, information
about how the poker chips link together may be valuable in
understanding the vessel structure as a whole. As discussed above,
the inventors have developed algorithms that facilitate linking
poker chips together to provide membership information with respect
to which poker chips belong to which vessel and information
regarding which poker chips are adjacent to one another. After
linking has been achieved, more sophisticated vessel analysis may
be performed.
[0091] Following below is a more detailed description of algorithms
capable of extracting geometry from 3D images to obtain a Poker
Chip.TM. representation of vasculature present in the images, in
accordance with some embodiments of the technology described
herein. While the various algorithms are discussed in connection
with detecting and extracting vessel information, the concepts
disclosed herein may be applied to detect and associate other
structure, as the aspects of the technology described herein are
not limited in this respect. In addition, it should be appreciated
that distribution analyses according to various aspects of the
technology described herein may be applied to information obtained
from any vessel image, representation, or combination thereof.
[0092] FIG. 3A illustrates one example of a cylindrical segment 300
that may be used to generally model a vessel segment. A
configuration of cylindrical segment 300 may be described by a
number of parameters in a particular coordinate frame. The position
of cylindrical segment 300 may be described by a location of the
cylindrical axis 305 at a point (x.sub.i, y.sub.i, z.sub.i) in
space, for example, the origin or termination of the cylindrical
segment. The orientation of cylindrical segment 300 may be
specified by the angle o.sub.i from the x-axis and the angle
.gamma..sub.i from the y-axis. Since cylindrical segment 300 is
axially symmetric, its rotation about the z-axis may not need to be
specified. The length of the cylindrical segment may be specified
by l.sub.i and the radius of the cylindrical segment 300 may be
specified by r.sub.i.
[0093] The inventors have appreciated that the cross-section of a
vessel may be characterized by a generally Gaussian shaped
intensity distribution. The cross-sectional density of a vessel may
be modeled by a Gaussian distribution, centered on the longitudinal
axis of the vessel, so that the modeled density is the highest at
the center of the vessel. For example, the cross-sectional density
distribution of a cylindrical vessel segment, when oriented such
that its longitudinal axis coincides with the z-axis, may be
modeled as,
.rho. ( 1 r 2 ( x - x i ) 2 + ( y - y i ) 2 ) ( 1 )
##EQU00001##
[0094] where .rho. is the density coefficient at a center of the
cylindrical segment and r is the radius of the cylindrical segment,
so that the density is modeled as being greatest at the center
(i.e., equal to .rho.) and decays exponentially as a function of
radial distance from the center. FIG. 3B illustrates a grey scale
representation of the function given in Eq. (1), where darker grey
scale values indicate increased density values. FIG. 3C illustrates
a plot of the intensity values along the x-axis at the center of
the grey scale Gaussian distribution in FIG. 3B. FIG. 3D
illustrates a vessel intensity profile that may better model the
intensity profile of vessels in an image. Curve 1 and 2 illustrated
vessel profile intensity when vessel diameter is larger than the
resolution of the scan and when the vessel diameter is smaller,
respectively.
[0095] The density distribution along the longitudinal axis of the
cylinder (i.e., into and out of the page in FIG. 3B) is
substantially uniform and does not vary substantially and may be
modeled as a constant function of the cross-sectional distribution
along the longitudinal axis, that is, as a constant function of the
radial distance d from the center of the distribution. FIG. 4
illustrates schematically a cylindrical vessel segment intensity
distribution model. In particular, the model of the cylindrical
vessel segment has a maximum density at the center that decays
exponentially to the boundary of the vessel as a function of the
radial distance d, from the center. At each distance d, the density
is uniform along the z-axis. For example, the density at d=0 is the
density maximum along the length of the vessel. This density
maximum shown by line 405 is referred to as a ridge, and
corresponds to the centerline of a vessel.
[0096] If the herein described characteristic intensity
distribution or similar distribution can be identified in the
image, the associated pixels/voxels are likely to belong to a
vessel. The characteristic points may be used to facilitate
segmenting the image into vessel and non-vessel regions. Some
methods of detecting the characteristic shape illustrated in FIG. 4
include performing ridge detection on an image. A ridge point is
defined herein as a point in an image wherein the intensity assumes
a local extrema in the direction of principal curvature, i.e., the
direction having the steepest intensity gradient. For example, at
point 415 (and along ridge 405) in FIG. 4, the principal direction
of curvature is shown by u.sub.0 (i.e., the unit vector (1,0) in
the (d, z) coordinate frame). Each point along ridge 405 forms a
ridge point since each point is a local maximum along the z-axis.
Accordingly, a ridge may be characterized by local derivative
information in the image and may be detected by examining the
curvature of intensity about points of interest in the image.
[0097] Some conventional methods have proposed detecting the ridge
using the Hessian operator. However, the Hessian operator requires
performing second derivatives of the image information, which
reduces the signal-to-noise ratio (SNR) and may result in degraded
performance. The inventors have developed methods of detecting the
characteristic shape of blood vessels described above using
centerline filtering techniques that may avoid some of the
performance degradations commonly seen with conventional filters
such as the Hessian operator, as discussed in further detail
below.
[0098] As discussed above in connection with FIG. 1, a non-limiting
example of a method for extracting geometry from images may include
a number of processing blocks including: a scale detector, an
orientation detector, centerline filtering, non-maximum suppression
and linkage. Briefly speaking, the system works as follows:
firstly, the scale detection and orientation detection modules may
be applied on 3D images to obtain correct size and orientation
parameters for centerline detection (e.g., scale and orientation
parameters for the centerline filters); secondly, based on the
parameters obtained from scale detection and orientation detection
modules, the centerline filter may be applied on every voxel of a
3D image, or applied on a subsection of voxels for which centerline
detection is desired. The generated response field formed by
applying the centerline filter indicates the likelihood that the
associated voxel corresponds to the vessel centerline; finally,
non-maximum suppression and linkage is applied on the centerline
response field to extract the vessel centerline and obtain a vessel
mathematical representation (e.g., a linked Poker Chip.TM.
representation). Following below are more detailed descriptions of
embodiments of the five main blocks briefly discussed above, e.g.,
scale detection, orientation detection, centerline filtering,
non-maximum suppression and centerline linking.
[0099] Scale Detection
[0100] As discussed above, scale detection may be applied to
estimate the centerline filter size appropriate for each voxel at
which centerline detection is to be applied. Applying scale
detection on each voxel of a 3D image volume may be relatively
expensive computationally. That is, if each voxel in the 3D image
is deemed to be a potential centerline point, then scale detection
should be applied to each voxel in the image. However, the
inventors have appreciated that since vessels occupy only a portion
of the volume, it may not be necessary to detect scale on every
voxel. In particular, certain voxels may be eliminated based on the
image properties of the voxels, for example, the intensity level of
the voxel.
[0101] In general, intensities from vessels are higher than those
in the background. Using a conservative intensity threshold, voxels
may be classified as background voxels with a low false positive
rate that can be controlled based on how conservative the threshold
operator is set. That is, by setting the threshold conservatively,
a substantial percentage of the background voxels may be eliminated
from scale detection without the risk of eliminating any vessel
voxels. The term "background" refers herein to voxels that are not
part of the subject matter of interest that is being detected. By
eliminating background voxels, the computations needed to perform
scale detection can be reduced. That is, by removing at least some
voxels from consideration, scale detection need not be performed on
each voxel in the image.
[0102] It is reasonable to model both background intensity and
vessel intensities as a Gaussian distribution. In practice, the
assumption in FIG. 5 shows that a model using a mixture of
truncated Gaussians is a very good fit for the data in low
intensity regions. The truncated Gaussian distribution has the
Probability Density Function (PDF) as follows:
p ( I .mu. , .sigma. ) = N ( I .mu. , .sigma. ) .intg. b 1 b 2 N (
x .mu. , .sigma. ) x ( 2 ) ##EQU00002##
[0103] where N(I|.mu., .sigma.) denotes a Gaussian distribution
with mean .mu. and variance .sigma., and b1 and b2 are the
truncation points. To capture both background and vessel
distributions, the mixture of two truncated Gaussians for the data
may be expressed as:
p ( I ) = c = 0 1 i { w c log [ N c ( I i .mu. c , .sigma. c )
.intg. b 1 b 2 N c ( x i .mu. c , .sigma. c ) x ] } ( 3 )
##EQU00003##
[0104] where w.sub.c is the weight percentage of each component.
Directly maximizing the likelihood may become challenging because
determining the marginal probability may require computations that
increase exponentially with the data. In some embodiments, the
problem is solved using an Expectation Maximization (EM) algorithm.
The EM process iteratively goes through two steps by soft
assignment of data (Expectation) and maximizing the whole
likelihood (Maximization). That is, an initial approximate
distribution may be used to classify voxels as either background or
foreground (e.g., vessels) in the Expectation step. Next, the
distribution is refined based on the classification (Maximization)
and classification (Expectation) is repeated on the refined
distribution. This process may be repeated until the process
converges on a final classification of background and foreground
voxels.
[0105] Applying an EM algorithm on a mixture of Gaussians is only
one method by which background voxels may be eliminated from
consideration, or by which voxels are classified as background and
foreground voxels. Other preprocessing or thresholding techniques
may be used to reduce the number of voxels on which further
processing is performed to reduce the computational expense, as the
aspects of the technology described herein are not limited in this
respect. In addition, while voxel intensity may be one suitable
parameter to use to perform a conservative elimination of voxels
belonging to the background, any suitable parameter may be used, as
the aspects of the technology described herein are not limited in
this respect. For example, higher order properties may be used.
[0106] As discussed above, separating scale detection and
orientation detection may have benefits over algorithms that
perform the two operations simultaneously. The inventors have
designed a scale detection filter which does not depend on the
orientation of the structure to be detected. According to some
embodiments, an orientation independent filter may be developed
such that the filter can be mathematically described in spherical
coordinates as f=f(r), which is a function that does not depend on
orientation. The symmetry of the filter allows the filter to be
independent of how the filter is oriented. To accurately detect
centerline voxels from 3D images, the response generated by the
scale detection filter should be maximum when it is located at a
centerline voxel. The scale .sigma..sub.r at a point (x, y, z)
inside a cylinder may be defined as the distance to the wall of the
cylinder boundary:
.sigma..sub.r(x,y,z)=dist(x,y,z; wall of the cylinder) (4)
[0107] As shown in FIG. 6, this definition of scale guarantees a
unique maximum filter response inside the cylinder after scale
selection (in the absence of noise). Normally, the intensity of a
3D image outside of a vessel is significantly lower than the
intensity inside the vessel. This rapid intensity decay provides an
indication of scale. The inventors have developed a rank-based
scale filter that is orientation independent. Given a point X
inside a vessel, a rank based scale filter may be defined as:
( X , r ) = f - ( { I ( X ' ) : X ' - X = r + 1 } ) min r { f + ( {
I ( X ' ) : X ' - X = 1 , , r } ) } ( 5 ) ##EQU00004##
[0108] where R(X, r) is the filter response at image location X
with filter radius r, and f.sub.- and f.sub.+ are rank functions,
respectively. Note that the filter is parameterized by radius only,
resulting in filter symmetry that is orientation independent. Given
various noise models, there are many ways to choose the rank
functions. In order to cope with image reconstruction effects,
f.sub.- may be chosen as the median value of the last 10 lowest
intensities and f.sub.+ may be chosen as the median value of the
last 10 highest intensities. That is, the rank function may be
determined from characteristics of the image. However, the rank
functions may be selected to be any value that facilitates
detection of scale, as the aspects of the technology described
herein are not limited in this respect. The scale .sigma..sub.r(X)
may then be obtained by finding the minimum radius r so that R(X,
r) reaches the threshold .alpha.:
.sigma. r ( X ) = min r { R ( X , r ) < 1 .alpha. } ( 6 )
##EQU00005##
[0109] Stated differently, the radius of the scale filter is
increased until the filter response no longer satisfies the
relationship in Eq. (6). As discussed above, the scale detection
filter may be designed to be independent of orientation. According
to some embodiments, the kernel or shell of the scale filter is a
circle in 2D and a sphere in 3D. As a result, the size of the
filter is defined by the radius r, where the center of the filter
is located at a target voxel at location X in the image. Since the
filter has the same radius in all directions, the application of
the scale filter is independent of orientation.
[0110] The criteria for the filter response may be chosen to be any
suitable criteria that can robustly determine when the filter
kernel has crossed a vessel boundary. The criteria in Eq. (6) is
merely exemplary. In some embodiments, the value of .alpha. is
chosen to be 5. However, other values may be used as well as the
aspects of the technology described herein are not limited in this
respect. In order to examine the sensitivities of this rank-based
scale filter to the choice of the threshold parameter .alpha., a
few points inside different vessels may be randomly chosen to see
how the selected scale changes depending on the ratio threshold
parameter .alpha.. FIG. 7 shows that the scale approaches the
correct value when .alpha. is chosen to be larger than 5.
[0111] FIG. 8 illustrates pictorial an orientation independent
scale filter, in accordance with some embodiments of the technology
described herein. It should be appreciated that while the scale
detection filter in FIG. 8 is shown (and is suitable) in the
context of a 2D image for convenience of illustration, the scale
detection filter is designed as a 3D filter to detect scale in 3D
volumetric images. In particular, the circular filter illustrated
in FIG. 8 may be made an expanded to a sphere to detect scale in
3D. In FIG. 8, a portion of an image 805 is shown having a vessel
structure 815 within the image portion. It should be appreciated
that image portion 805 is schematic and the vessel structure 815
and the background 825 would be comprised of an intensity value at
each voxel location in the image portion. Moreover, it should be
appreciated that image portion 805 may be a small portion of a much
larger image. For the sake of clarity only a single vessel
structure is depicted in image portion 805, though the image
portion may in reality include any number of vessel structures.
[0112] FIG. 8 also illustrates three separate applications of an
orientation independent scale filter 850. It should be appreciated
that the scale filter 850 may be applied at all of the image voxels
or at a selected number of image voxels (e.g., voxels determined to
be vessel voxels using a preprocessing techniques such as the
intelligent thresholding method described above). The three
applications of the filter in FIG. 8 are merely exemplary and are
chosen at arbitrary locations to assist in describing the scale
detection filter. Each application of the filter begins by placing
the filter with a predetermined minimum radius r on a target pixel
at which scale is being detected. The scale filter is then applied
to the image, for example, by convolving the image pixels that fall
under the filter kernel or support with the values of the filter
kernel. If a certain criteria is met, the filter is assumed to
still be entirely within the vessel and the radius r is
increased.
[0113] In FIG. 8, the increasing of the filter radius is depicted
by the successively larger circles in dashed line. The circles in
solid line denote the last filter applied such that the criteria
was met. For example, the dotted line circle in filter application
850b shows a circle of r.sub.n that when applied to the underlying
image failed to meet the criteria, where n is the number of
successively larger radius filter kernels that have been applied to
the image. Thus, the scale at the corresponding image location is
determined to be r.sub.n-1. Not only does scale detection provide
the appropriate scale to be used in subsequent filtering processes
(e.g., centerline detection), it also may indicate the radius of
the vessel structure in the Poker Chip.TM. representation.
[0114] The inventors have used the fact that the intensity of
voxels within the vessel, in the absence of noise, is substantially
higher than the background voxels to establish the criteria such
that the criteria will not generally be met when the filter kernel
is extended outside the vessel structure. One embodiment of such a
criteria is described in Eq. 5 and Eq. 6. By employing the rank
functions illustrated in Eq. 5, and using the criteria in Eq. 6, a
robust filter may be designed that will fail to meet the criteria
when the filter kernel is increased in size such that it
encompasses voxels outside of the vessel. However, the above
described scale detection filter is exemplary and other scale
detection filters may be used, as the aspects of the technology
described herein are not limited in this respect. In addition, any
criteria that tends not to be met as a filter is expanded across a
vessel boundary may be used, as the aspects of the technology
described herein are not limited in this respect.
[0115] Because the centerline voxels are not known a priori, the
scale detection filter may be applied to non-centerline voxels. As
shown by filter application 850b, the scale detection is again
stopped when the filter kernel crosses the vessel boundary. Because
the target voxel is not a centerline voxel, the radius of the
filter will not correspond to the radius of the vessel. However,
this may be inconsequential because voxels that are not determined
to be centerline voxels are removed in subsequent processing, such
as during centerline filtering discussed below. Because only voxels
detected as centerline voxels will survive centerline filtering,
the radius of the scale detector may accurately reflect the radius
of the associated vessel.
[0116] FIG. 9 shows what R(X, r) looks like when it is applied on
real images. Although the intensities have large variation inside
the vessel, the rank-based scale filter behaves smoothly and decays
relatively rapidly across the boundary of the vessel. Thus,
rank-based scale filters may have the generally beneficial property
of relatively distinct response change as the filter crosses vessel
boundaries, and is relatively stable and insensitive to the choice
of ratio parameter. Accordingly, scale may be detected at each
selected voxel in the image. For example, scale may be detected at
each voxel in the image or the reduced number of voxels resulting
from performing thresholding on the image to eliminate at least
some of the background voxels. The selected voxels at which scale
detection is performed can be selected in other ways, as the
aspects of the technology described herein are not limited in this
respect.
[0117] Orientation Detection
[0118] As discussed above, centerline filtering may be improved by
first determining the orientation at which the centerline filter
should be applied. Since scale is detected independent of
orientation, orientation detection may be performed separately from
scale detection and, in some embodiments, orientation detection
uses the scale values detected during scale detection to improve
detection of the orientation of the subject matter of interest. In
some embodiments, a gradient based orientation detection algorithm
may be used, however, other algorithms may be used to detect vessel
orientation, as the aspects of the technology described herein are
not limited in this respect. Because of the rotational symmetry
along the axis of a cylinder on which the vessel structure may be
modeled, the intensity along a line parallel to the vessel axis is
constant in the absence of noise. In other words, the directional
derivative of intensity along the direction v parallel to the
vessel axis is zero in the absence of noise:
v.gradient..rho.(X)=0 (7)
[0119] It should be appreciated that x-ray decay during image
acquisition depends on its penetrating length. Thus, the intensity
inside a vessel tends to vary along any direction other than the
axis direction. This fact indicates that Eq. (7) may be a necessary
and sufficient condition for finding the vessel direction since the
above argument holds for any point X inside the vessel. Therefore,
the direction of a small cylinder segment at each point X can be
estimated by finding a direction vector a along which the
intensities have the least change. However, direct estimation from
the derivative of one point X tends to be error prone. In some
embodiments, all the derivatives inside a small volume centering on
the point X may be used to increase the accuracy. To be more
precise, the axis direction a may be estimated by finding a
direction a that minimizes the sum of the directional intensity
gradient along this direction:
a ^ = arg min a { .intg. .intg. v .intg. a .gradient. .rho. ( x , y
, z ) x y z } ( 8 ) ##EQU00006##
[0120] where .sigma.(X) is the scale detected at point X and
.parallel..parallel. is the norm discussed herein. In the presence
of noise, a directional gradient of intensity convolved with an
adaptive Gaussian kernel may be used, as follows.
a ^ = arg min a { .intg. .intg. v .intg. a .gradient. ( G .sigma. (
x , y , z ) .smallcircle. .rho. ( x , y , z ) ) x y z } ( 9 )
##EQU00007##
[0121] In some embodiments, Eq. (9) can be solved by a least square
estimation by assuming the noise distribution is Gaussian i.i.d,
i.e., the norm in Eq. (9) is an L2-norm. However, it is well known
that an L2-norm may be sensitive to outliers present in the input
data, and outliers may frequently appear in reconstructed 3D
images. In some embodiments, a L1-norm in Eq. (9) may be used.
a ^ = arg min a { .intg. .intg. v .intg. a .gradient. ( G .sigma. (
x , y , z ) .smallcircle. .rho. ( x , y , z ) ) 1 x y z } ( 10 )
arg min a { .intg. .intg. v .intg. a 1 .gradient. ( G .sigma. ( x ,
y , z ) .smallcircle. .rho. ( x , y , z ) ) 1 x y z } ( 11 )
##EQU00008##
[0122] To avoid the trivial solution at a=0 in the above equation,
the constraint .SIGMA..sub.1.parallel.a.sub.1.parallel..sub.2=1 may
be used. Since a is independent of the point (x, y, z), a is moved
out of the triple integral so that:
a ~ = min a { a .intg. .intg. v .intg. .gradient. ( G .sigma. ( x ,
y , z ) .smallcircle. .rho. ( x , y , z ) ) x y z M L 2 } s . t . {
i a i 2 = 1 } ( 12 ) ##EQU00009##
[0123] It should be appreciated that in Eqs. (8)-(12), the
operation is being performed over a volume v. By performing
orientation detection over a neighborhood, rather than at a single
voxel, semi-global information may be captured in the orientation
assessment. The neighborhood information allows for robust
orientation detection in the presence of noise and outliers.
However, it should be appreciated that the neighborhood (e.g., the
volume v) may be different for detecting direction in relatively
large vessels versus relatively small vessels. Accordingly, the
inventors have developed an adaptive method that varies the size of
the neighborhood based on the scale at a target voxel. That is, the
scale determined during scale detection may be used to determine
the size of the volume v. In some embodiments, the size of (2.left
brkt-bot.s+2.right brkt-bot.+1) may be used as the size of volume.
However, any adaptive neighborhood based on scale may be used, as
the aspects of the technology described herein are not limited in
this respect. Thus, the size of the neighborhood used for
orientation detection may be adapted according to the scale of the
image at each location.
[0124] As discussed above, and L1-norm may be used to address
outliers. There are a number of ways to solve Eq. (12). In some
embodiments, the equation is solved by constraint optimization
using Lagrange multipliers. Applying Lagrange multipliers to the
above equation obtains:
.gradient..sub.a(a.sup.TM.sup.TMa+.lamda.a.sup.Ta)=0
(M.sup.TM)a+.lamda.a.sup.T=0 (13)
[0125] Therefore the center line direction, a, may be obtained by
computing the eigenvector associated with the smallest eigenvalues
of matrix M. Referring back to FIG. 4, solving the above equations
to determine the direction a can be pictorial explained. In general
terms, the eigenvectors of matrix M indicate the characteristic
directions of curvature. The relationship between these
characteristic directions of curvature may be employed to identify
the direction of the centerline. The eigenvalues and associated
eigenvectors of a matrix may be determined in various ways, for
example, by any number of well-known iterative methods of
diagonalizing a matrix or analytically by directly solving the
relationship:
Mu=.lamda.u (14)
[0126] where M is the matrix of Eq. 13, u is an eigenvector of
matrix M, and .lamda. is an eigenvalue associated with u. The
magnitude of each eigenvalue of the matrix M is related to the
"significance" of the associated eigenvector. Stated differently,
the eigenvalue indicates how much the curvature along the
associated eigenvector contributes to the local curvature
determined by the matrix M. Accordingly, a in Eq. 13 is the
eigenvector associated with the smallest eigenvalue and indicates
the direction in which the change in intensity is the smallest. The
largest eigenvalue of the matrix M is associated with the principal
direction of curvature.
[0127] In FIG. 4, the linearly independent eigenvectors u.sub.0 and
u.sub.1 (i.e., eigenvectors u.sub.0 and u.sub.1 are orthogonal) are
shown on the illustrated intensity curve. The eigenvalue
.lamda..sub.0 herein denotes the eigenvalue having the greatest
absolute value and is referred to as the principal eigenvalue.
Accordingly, the associated eigenvector u.sub.0 indicates the
principal direction of curvature at a target pixel and
.lamda..sub.0 is related to the magnitude of the curvature. The
eigenvalue .lamda..sub.1 (referred to as the secondary eigenvalue)
is related to the magnitude of curvature in the direction of
u.sub.1, i.e., in a direction orthogonal to the principal direction
of curvature indicated by u.sub.0. Along the ridge of the Gaussian
profile (i.e., in the direction u.sub.1), the intensity should be
substantially zero and the change in intensity relatively small and
in the noiseless case is zero (i.e., the intensity does not change
as a function of z in the direction of the centerline).
Accordingly, by determining the eigenvector associated with the
smallest eigenvalue, the direction a which corresponds to the
direction of the centerline may be determined. Thus, the
orientation of the centerline may be determined at each of the
selected voxels.
[0128] Centerline Detection
[0129] Having determined scale and orientation for the feature
detection filter, the feature of interest may be detected.
According to some embodiments, centerline detection is performed
using a Gaussian centerline filter. For example, assume the density
inside the vessel satisfies the Gaussian model:
I ( r ) = I 0 e - r 2 2 .sigma. 2 ( 15 ) ##EQU00010##
[0130] Here, r is in the direction perpendicular to the vessel
axis; .sigma. is the radius of the vessel; and I.sub.0 is the
intensity at the center. In order to detect a Gaussian vessel, a
filter with radial variation corresponding to the 2nd derivative of
the Gaussian may be used:
h ( r ) = ( r 2 .sigma. 2 - 1 ) - r 2 .sigma. 2 ( 16 )
##EQU00011##
[0131] The application of this filter corresponds to a volume
integral over space. This volume integral should vanish if the
filter is embedded in material with constant density. However the
2nd derivative of the Gaussian does not, i.e.,
.intg. 0 .infin. ( r 2 .sigma. 2 - 1 ) - r 2 .sigma. 2 r r = 1 ( 17
) ##EQU00012##
[0132] This problem can be fixed by adding an offset,
.intg. 0 .infin. ( r 2 .sigma. 2 - 2 ) r 2 .sigma. 2 r r = 0 ( 18 )
##EQU00013##
[0133] Therefore, the centerline filter has the form
f ( r ) = e 4 .PI. .sigma. 2 [ 2 - [ r .sigma. ] 2 ] r 2 2 .sigma.
2 ( 19 ) ##EQU00014##
[0134] This filter has a positive core when r< {square root over
(2)}.sigma.r< and negative shell when r> {square root over
(2)}.sigma..
[0135] The inventors have appreciated that in the presence of
noise, a centerline filter that closely mimics the shape of a
Gaussian as described above may at times be inaccurate, especially
in situations where vessel structures are relatively close
together. In particular, the continuous decay of the Gaussian may
incorrectly detect or fail to detect centerline voxels in certain
situations, such as when vessel structures are close together
and/or in circumstances where relatively small vessel structures
appear nearby relatively large vessel structures.
[0136] The inventors have appreciated that a modified centerline
filter may be more effective at accurately identifying centerline
points, particularly in the presence of noise. According to some
embodiments, centerline detection is performed using a filter that
better matches the profile of vessel structures in an image. FIG.
10A illustrates a matched filter in accordance with some
embodiments of the technology described herein. Filter 900 includes
an inner core and an outer core. Rather than a Gaussian kernel,
filter 900 includes a step function between the inner and outer
core. As a result, the filter support is more compact and the
filter is able to more accurately detect vessel structures that are
close together. In addition, because the filter better matches
vessel profiles, centerline detection may be more accurate. An
example of values assigned to the matched filter 900 according to
some embodiments include:
f s ( r , z ) = { 1 r .ltoreq. s and z .ltoreq. 2 s 0 s < r
.ltoreq. 2 s and z .ltoreq. 2 s - 1 r > 2 s or z > 2 s ( 20 )
##EQU00015##
[0137] An illustration of the profile of the above filter along the
axis x-x' is shown pictorially in FIG. 10B. As shown, the size of
the matched filter is based on the scale s detected during scale
detection. Applying this filter, the centerline response may be
given as:
r(x,y,z)=.intg..intg..intg.T[f(r,z)G(0,.sigma.]1(x,y,z)dxdydz
(21)
[0138] where G(0, .sigma.) is a Gaussian smooth kernel. When the
scale of the filter is small (e.g., when scale detection determines
that the local scale is relatively small), the filter defined by
Eq. (20) may not have a zero net volume (volume of the positive
core minus the volume of the negative core). This may cause
detection difficulties because the filter may have non-zero
response when applied to a non-zero uniform background. As shown in
the FIG. 12, when the scale of the filter is small, the net volume
percentage may be quite large. For example, for a centerline filter
with scale of 1.5, the net volume is 35% of the total volume of the
filter. Thus, the filter may generate filter bias in the favor of
small scale.
[0139] Therefore, to address this bias the filter described above
may be modified as:
f s ( r , z ) = { 1 r .ltoreq. s and z .ltoreq. 2 s 0 s < r
.ltoreq. .sigma. ( s ) and z .ltoreq. 2 .sigma. ( s ) - w s r >
.sigma. ( s ) or z > 2 .sigma. ( s ) where , ( 22 ) .sigma. ( s
) = { 2 s + 0.5 if s < 10 2 s otherwise ( 23 ) ##EQU00016##
and w.sub.s is a function of scale s so that,
.intg..intg..intg..sub.r>.sigma.(s) or z> {square root over
(2)}.sigma.(s)w.sub.sdxdydz=.intg..intg..intg..sub.r.ltoreq.s and
z.ltoreq. {square root over (2)}sdxdydz (24)
[0140] An illustration of the profile of the filter expressed in
Eq. (22) along the axis x-x' is shown pictorially in FIG. 10C. The
matched filters described above may be particularly effective at
accurately detecting centerline voxels in the presence of noise and
in circumstances when subject matter of interest is positioned in
close proximity to each other.
[0141] The matched filters described above may be applied to a
plurality of selected voxels in the image. Accordingly, for each
selected voxel at which the matched filter is applied, there will
be an associated filter response indicative of the likelihood that
the corresponding voxel is a centerline voxel. However, only the
maximum filter responses may be of interest. That is, the maximum
filter responses are those that are most likely to be centerline
voxels. Accordingly, filter responses that are not maximum may be
suppressed such that only those voxels having maximum filter
responses remain.
[0142] Non-Maximum Suppression
[0143] In some embodiments, non-maximum suppression may be
performed. For example, after centerline filtering, each voxel has
a response. The response on each voxel indicates how likely it is
that the voxel is a centerline voxel. Since the center line voxel
should have the maximum response in the plane perpendicular to the
axis, the purpose of non-maximum suppression is to suppress
non-maximum responses to eliminate non-centerline voxels. On each
voxel, a cutting plane perpendicular to the vessel axis may be used
to suppress the non-maximum responses. On the cutting plane, only
local maximums of centerline filter responses are kept and all
other responses are suppressed. Interpolating the centerline
location in order to achieve sub-voxel accuracy is described
below.
[0144] In some embodiments, location interpolation on the cutting
plane may be performed. After obtaining the direction of the
cylinder, a cutting plane perpendicular to this direction may be
used to apply the non-maximum suppression as an analog to the
traditional computer vision edge detection problem. Given an
arbitrary voxel x, the voxel x may be tested to determine whether
the voxel is a local maxima. According to some embodiments, the
cutting plane may be centered on x and the centerline response R(x)
may be compared with any other responses in its cutting plane
neighborhood N(x, v.sub.x). That is, the response field in the
neighborhood N (e.g., a 3.times.3.times.3 neighborhood) may be
projected onto this cutting plane. If the response at voxel x is
larger or equal to all of the responses of neighborhood voxel,
voxel x may be labeled as a local maxima. Otherwise, voxel x is
labeled as a non-maxima voxel and suppressed. This test may be
expressed as:
IsMaxima ( x ) = { true R ( x ) .gtoreq. R ( y ) , .A-inverted. y
.di-elect cons. ( x , v z ) false otherwise ( 25 ) ##EQU00017##
[0145] where N(x,vx) denotes the cutting plane neighborhood of the
point x. Once the neighborhood is determined, the parabolic
function as shown below may be used to interpolate the sub-voxel
maximum location.
r(x,y)=ax.sup.2+by.sup.2+cxy+dx+ey+f (26)
[0146] Given the above response model and the centerline filter
responses in a small region around the center, the following
equations may be used:
an 2 + bm 2 + cmn + dn + em + f = r ( n , m ) a ( n - 1 ) 2 + bm 2
+ cm ( n - 1 ) + d ( n - 1 ) + em + f = r ( n - 1 , m ) a ( n - 1 )
2 + bm 2 + cm ( n - 1 ) - d ( n - 1 ) - em + f = r ( 1 - n , - m )
an 2 + bm 2 + cmn - dn - em + f = r ( - n , - m ) ( 27 )
##EQU00018##
[0147] This linear form can be written as a matrix form
A [ a b c d e f ] = [ r ( n , m ) r ( n - 1 , m ) r ( 1 - n , - m )
r ( - n , - m ) ] ( 28 ) where A = [ n 2 m 2 mn n m 1 ( n - 1 ) 2 m
m ( n - 1 ) n - 1 m 1 n 2 m 2 m ( n - 1 ) 1 - n - m 1 n 2 m 2 mn -
n - m 1 ] ( 29 ) ##EQU00019##
[0148] The maximum location is determined by the stationary
condition
.differential. r .differential. x = .differential. r .differential.
y = 0. ##EQU00020##
That is,
[0149] 2ax+cy_d=0
cx+2by+e=0 (30)
Therefore,
[0150] [ x y ] = - [ 2 a c c 2 b ] - 1 [ d e ] = 1 4 ab - c 2 [ - 2
b c c - 2 a ] [ d e ] = [ ce - 2 bd 4 ab - c 2 cd - 2 ae 4 ab - c 2
] ( 31 ) ##EQU00021##
[0151] In some embodiments, the size of the neighborhood N(x,vx) is
determined based characteristics of the image in the neighborhood.
There is a natural question of how big the neighborhood size should
be chosen in the non-maximum suppression algorithm. In some
embodiments, the smallest size of 3.times.3.times.3 may be used,
but this choice may cause outliers to survive non-maximal
suppression in noisy regions. An alternative method of choosing the
parameter is to use the results from radius and/or scale detection.
In some embodiments, to avoid suppressing real vessels which are
close to each other, a conservative approach may be used when
choosing the neighborhood:
n = 2 s 2 = 1 ( 32 ) ##EQU00022##
[0152] It should be appreciated that the neighborhood in Eq. (32)
is exemplary and an adaptive neighborhood, for example, based on
scale may be determined in other ways, as the aspects of the
technology described herein are not limited in this respect.
[0153] Generating a Linked Representation of a Vessel Network
[0154] As discussed above, information obtained from one or more
images of a vessel network may be used to generate an unlinked
representation of the vessel network. The unlinked representation
may comprise one or more geometric objects (e.g., Poker Chips.TM.)
each of which represents a cross-section of a vessel segment in the
vessel network. Each of the geometric objects may represent a
centerline voxel (e.g., when the geometric object is a Poker
Chip.TM., the center location of the Poker Chip.TM. corresponds to
a centerline voxel). An unlinked representation of the vessel
network may be obtained based on output from centerline filtering
and non-maximum suppression processes, which provide a 3D field in
which each point is marked as either belonging to or not belonging
to a centerline. The centerline points may be associated with other
information such as radius, strength and orientation (e.g., using
the Poker Chip.TM. representation).
[0155] However, without further processing, an unlinked
representation of a vessel network does not by itself provide a
notion of adjacency or vessel membership, which may be useful in
performing analysis of vessel structure in the vessel network.
Accordingly, in some embodiments, an unlinked representation of a
vessel network may be processed to generate a linked representation
of the vessel network. The linked representation may comprise
information indicating structure of individual vessel segments
(e.g., what centerline voxels belong to which vessel segments) as
well as how the vessel segments connect to one another.
[0156] In some embodiments, the linked representation comprises a
plurality of geometric objects (e.g., Poker Chips.TM.) along with
information indicating how the geometric objects are linked to
create linked representations of vessel segments in the vessel
network. In some embodiments, a linked representation of a vessel
network comprises a vessel network structure graph that represents
the connectivity of vessel segments in the vessel network. An edge
in the vessel network structure graph may represent a vessel
segment and a vertex in a vessel network structure graph may
represent an intersection of two vessel segments in the vessel
network (e.g., at a branch point).
[0157] FIG. 20 is a flowchart of illustrative process 2000 for
generating a linked representation of a vessel network, in
accordance with some embodiments. Process 2000 may be executed
using any suitable system and, for example, may be executed using
computer system 2800 described below with reference to FIG. 28.
[0158] Process 2000 begins at act 2002 where an unlinked
representation of a vessel network is obtained. For example, in
some embodiments, a geometry comprising a plurality of geometric
objects may be obtained, where each geometric object (e.g., a Poker
Chip.TM.) represents a cross-section of a vessel segment in the
vessel network, as described above. The geometric objects may be
obtained in any suitable way (e.g., the Poker Chips.TM. may be
obtained in any of the ways described herein).
[0159] Next, process 2000 proceeds to act 2004, where a linked
representation of a vessel segment is generated. In some
embodiments, an initial geometric object (e.g., a prominent Poker
Chip.TM., such as a Poker Chip.TM. with a relatively large radius)
is identified and one or more other geometric objects are linked to
the initial geometric object to form a linked representation of a
vessel segment, whereby the linked geometric objects represent
cross-sections of the vessel segment. The linking may be performed
in accordance with process 2100 described below with reference to
FIG. 21, or in any other suitable way.
[0160] Next, process 2000 proceeds to act 2006, where the linked
representation (of the vessel segment) obtained at act 2004 is
further processed to identify branch points on the vessel segment.
This may be done in any of the ways described below including, but
not limited to, using processes 2200, 2300 and 2500 described below
with reference to FIGS. 22, 23, and 25, respectively. Identifying a
branch point on the vessel segment may comprise identifying the
location of the branch point on the segment as well as identifying
the type of the branch point.
[0161] Next, process 2000 proceeds to act 2008, where one or more
geometric objects that represent a vessel branch (i.e., another
vessel segment branching off of the vessel segment whose
representation was analyzed at act 2006 to identify one or more
branch points) are identified for each of one or more of the branch
points identified at act 2006. In some embodiments, one or more
unlinked geometric objects may be identified (e.g., one or more
Poker Chips located in a neighborhood of a branch point and having
large radii) as representing a vessel branch. In some embodiments,
one or more linked geometric objects providing a linked
representation of the vessel branch may be identified (such a
representation may be produced, for example, as part of the process
for identifying a branch point, as described in more detail below).
In any case, the geometric objects that represent a vessel branch
may be used subsequently to perform further processing on the
vessel branch (e.g., perform linking or further linking, identify
branch points etc.).
[0162] Next, process 2000 proceeds to act 2010, where the vessel
network structure graph may be updated based on results of acts
2004-2006. In some embodiments, the graph may be updated to have a
vertex for each branch point identified at act 2006 and an edge
between any two vertices representing branch points connected by a
single vessel segment. For example, if a single vessel segment were
identified, at act 2006, as having M consecutive branch points
(where M is an integer greater than or equal to 1), the graph may
be updated to have a vertex for each of the M branch points and an
edge between vertices representing branch points that are adjacent
on the vessel segment.
[0163] Next, process 2000 proceeds to decision block 2012 where it
is determined whether another segment (e.g., one or more segments
branching off of the vessel segment processed at acts 2004-2010) is
to be processed further. This determination may be made in any
suitable way. For example, it may be determined that another
segment is to be processed when a segment branching off of the
vessel segment has not been itself processed to identify branch
points thereon. When it is determined at decision block 2012 that
no other segment is to be processed, process 2000 completes.
[0164] On the other hand, when it is determined that another
segment (e.g., a branch of the vessel segment analyzed at acts
2004-2010) is to be processed, a representation of the segment to
be processed is obtained (e.g., one or more geometric objects
representing the segment to be processed as may have been obtained
at act 2008), and process 2000 returns via the "YES" branch to act
2004, at which point acts 2004-2010 are repeated. The segments to
be processed may be identified in any suitable way. In some
embodiments, each branch of a vessel segment is selected to be
processed before any branch of a branch of the vessel segment is
selected to be processed--a breadth-first-search type approach. In
some embodiments, a branch of a vessel and all its sub-branches are
selected to be processed before any other branch of the vessel
segment is selected--a depth-first search type approach.
[0165] Linking
[0166] As discussed above, generating a representation of a vessel
network may comprise generating a linked representation of one or
more vessel segments in the vessel network by linking centerline
voxels (e.g., center locations of Poker Chips.TM.) to identify
which centerline voxels are adjacent and determine the vessel
segments to which the centerline voxels belong. Accordingly, in
some embodiments, generating a linked representation of a vessel
segment may be performed by linking geometry extracted from one or
more images (e.g., obtained via CT scan, Magnetic Resonance
Imaging, Optical Coherence Tomography, etc.). The geometry may
include a plurality of geometric objects (e.g., Poker Chips.TM.)
each of which represents a cross-section of a vessel segment
Linking the geometric objects, by associating groups of the
geometric objects with vessels and determining which geometric
objects are adjacent to one another, provides a linked
representation of one or more vessel segments in the vessel
network. Each of the geometric objects may represent a centerline
voxel (e.g., when the geometric object is a Poker Chip.TM., the
center location of the Poker Chip.TM. corresponds to a centerline
voxel).
[0167] In some embodiments, the geometric objects may be linked
according to a criteria that includes one or any combination of
minimizing a distance, a direction change, a radius change, and/or
a filter response change from a geometric object to an adjacent
geometric object. That is, when selecting between a number of
candidate geometric objects to link to a target geometric object,
the geometric object candidate that creates the smallest change in
one or more of the above parameters may be preferred over candidate
centerline geometric objects creating larger changes.
[0168] Accordingly, in some embodiments, each of the geometric
objects in the geometry extracted from one or more images may be
associated with one or more parameter values. In turn, these
parameter values may be used to determine how to link the geometric
objects to produce a linked representation of a vessel segment. For
example, each geometric object may be associated with one or more
location values indicative of its location (e.g., a Poker Chip.TM.
may be associated with one or more values indicating the location
of its center, which corresponds to a centerline voxel), and the
location values of geometric objects may be used to determine how
they should be linked (e.g., geometric objects closer to one
another are more likely to be linked than geometric objects that
are farther apart). As another example, each geometric object may
be associated with one or more direction/orientation values (e.g.,
a Poker Chip.TM. may be associated with one or more values
indicating the orientation of the centerline at the location of the
Poker Chip.TM.), and the direction/orientation values of geometric
objects may be used to determine how they should be linked (e.g.,
geometric objects with similar direction/orientation parameter
values are more likely to be linked than geometric objects having
disparate direction/orientation parameter values). As yet another
example, each geometric object may be associated with one or more
scale values (e.g., a Poker Chip.TM. may be associated with one or
more values indicating the radius/diameter of the Poker Chip.TM.),
and the scale values of geometric objects may be used to determine
how they should be linked (e.g., geometric objects of similar scale
are more likely to be linked than geometric objects having
disparate radii). As yet another example, each geometric object may
be associated with one or more response values corresponding to the
response of a scale detection filter used to determine the scale of
the geometric object, and the response values of geometric values
may be used to determine how the geometric objects should be linked
(e.g., geometric objects having similar response values are more
likely to be linked than geometric objects having disparate
response values). It should be appreciated that geometric objects
are not limited to having only the above-described parameter values
and may be associated with one or more values of any other suitable
parameters, as aspects of the technology provided herein are not
limited in this respect.
[0169] In some embodiments, all of the parameter values associated
with geometric objects may be used to determine how the geometric
objects are to be linked. In other embodiments, only some of the
parameter values (e.g., location parameter values only;
direction/orientation parameter values only; scale parameter values
only; response parameter values only; location and
direction/orientation parameter values only; location and scale
parameter values only; location and response parameter values only;
direction/orientation and scale parameters; direction/orientation
and response parameters only; scale and response parameters only;
location, direction/orientation and scale parameter values only;
location, direction/orientation, and response parameter values
only; location, scale and response parameter values only,
direction/orientation, scale, and response parameter values only)
may be used to determine how the geometric objects are to be
linked, as aspects of the technology described herein are not
limited in this respect.
[0170] FIG. 21 is a flowchart of illustrative process 2100 for
linking geometric objects to generate a linked representation of a
vessel segment. Process 2100 may be executed using any suitable
system and, for example, may be executed using computer system 2800
described below with reference to FIG. 28. Process 2100 may be
performed after a plurality of geometric objects (e.g., Poker
Chips.TM.) are obtained from one or more images using any of the
techniques described herein.
[0171] Process 2100 begins at act 2102, where one of the plurality
of geometric objects is identified as an initial geometric object
to be used in generating a linked representation of a vessel
segment. The initial geometric object may be identified from among
the plurality of geometric object to be the most prominent
geometric object. For example, the initial geometric object may be
identified as the geometric object in the plurality of geometric
objects having the largest scale (e.g., radius) and/or the largest
response. Though, the initial geometric object may be identified in
any other suitable way, as aspects of the technology described
herein are not limited in this respect. An initial direction in
which to search for candidate geometric objects to link to the
initial geometric object may be set equal to (or opposite to) the
direction/orientation of the initial geometric object. The initial
geometric object and the initial direction are then set as the
target geometric object and the target direction, respectively, at
act 2104 of process 2100.
[0172] Next, process 2100 proceeds to act 2106, where a geometric
object to link to the target geometric object is selected from
among multiple candidate geometric objects (e.g., at least two, at
least five, at least ten, at least twenty five, at least one
hundred, at least five hundred, at least one thousand, at least ten
thousand candidate geometric objects). As discussed above, in some
embodiments, the selection may be performed by comparing parameter
values of the target geometric object with parameter values of the
candidate geometric objects. Any of the above-described parameter
values or any suitable combination of them may be used to perform
the comparison. For example, in some embodiments, the selection may
be performed at least in part by comparing at least one value for
location of the target geometric object to respective values for
location of the candidate geometric objects, and comparing at least
one value for direction/orientation of the target geometric object
to respective values for direction/orientation of the candidate
geometric objects. Additionally or alternatively, the selection may
be performed by comparing at least one value for scale of the
target geometric object to respective values for scale of the
candidate geometric objects. Additionally or alternatively, the
selection may be performed by comparing at least one value for
response of a scale detection filter (e.g., the scale detection
filter used to detect scale of centerline voxels in any of the ways
described herein) of the target geometric object to respective
values for response of the scale detection filter of the candidate
geometric objects.
[0173] In some embodiments, comparing parameter values of the
target geometric object with respective parameter values of a
candidate geometric object is performed by using a statistical
model that provides, based on parameter values of the target and
candidate geometric objects, a likelihood that the candidate
geometric object and the target geometric object each represent
cross sections of the same vessel segment. The statistical model
may provide a likelihood that the candidate geometric object
follows the target geometric object as a representation of another
cross section of the same vessel segment. The statistical model may
be used to obtain a likelihood that the target and candidate
geometric objects each represent cross sections of the same vessel
based on some (e.g., all) parameter vales of the target and
candidate geometric objects. As one non-limiting example, the
statistical model may provide the likelihood based at least in
part, on at least one location value and at least one
direction/orientation value of the target geometric object and at
least one location value and at least one direction/orientation
value of the candidate geometric object. As another non-limiting
example, the statistical model may provide the likelihood based, at
least in part, on at least one location value, at least one
direction/orientation value, at least one scale value and at least
one response value for each of the target and candidate geometric
objects.
[0174] In some embodiments, the statistical model may be used to
calculate a likelihood likelihood that the candidate geometric
object follows the target geometric object by calculating a
likelihood/probability of parameters of the candidate geometric
object conditioned on parameters of the target geometric object.
This may be done in any suitable way, one non-limiting example of
which is described in more detail below.
[0175] In one non-limiting embodiment the statistical model may
comprise a probability distribution representing the probability
that target geometric object x (which is associated with location
parameter value(s) l.sub.x, direction/orientation parameter
value(s) v.sub.x, scale parameter value(s) s.sub.x, and response
parameter value(s) r.sub.x) and candidate geometric object y (which
is associated with location parameter value(s) l.sub.y,
direction/orientation parameter value(s) v.sub.y, scale parameter
value(s) s.sub.y, and response parameter value(s) r.sub.y)
represent cross sections of the same vessel segment. It should be
appreciated that direction/orientation v.sub.x is the direction of
the target geometric object set at one of acts 2104, 2114, or 2216.
We denote this distribution as:
Pr(L.sub.y=x|l.sub.x,v.sub.x,s.sub.x,r.sub.x). (33)
[0176] This expression may be viewed as a posterior probability
distribution with respect to the candidate geometric object. That
is, when the target geometric object x is fixed, the posterior
probability distribution will evaluate to a different value for
every candidate target geometric object under consideration.
Accordingly, in some embodiments, selecting a candidate geometric
object to link to the target geometric object x may be performed by
evaluating the probability distribution of Eqn. 33 for each of two
or more candidate objects (holding the target candidate object
fixed) and then selecting the candidate geometric object having the
highest probability (according to Eq. 33) among those candidate
geometric objects evaluated.
[0177] Absent assumptions on the prior distribution(s) of the
variables l.sub.x, v.sub.x, s.sub.x, and r.sub.x, maximizing the
probability distribution of Eqn. 33 is equivalent to maximizing the
likelihood
Pr(l.sub.x,v.sub.x,s.sub.x,r.sub.x|L.sub.y=x) (34)
[0178] Under certain independence assumptions, the likelihood
function of Eqn. 34 may be factored into a product of
lower-dimensional distributions according to:
Pr ( l x , v x , s x , r x L y = x ) = Pr ( dist ( x , y ) , xy
.fwdarw. , s , s x , s y , r x , r y L y = x ) = Pr ( dist ( x , y
) l x ) Pr ( xy .fwdarw. v x ) Pr ( r y , s y r x , s x ) , ( 35 )
##EQU00023##
where the probability distribution Pr(dist(x, y)|l.sub.x)
represents the probability that candidate geometric object y and
target geometric x represent cross sections of the same vessel
segment based on location parameter values of the target and
candidate geometric objects, where the probability distribution
Pr({right arrow over (xy)}|v.sub.x) represents the probability that
candidate geometric object y and target geometric x represent cross
sections of the same vessel segment based on direction/orientation
parameter values of the target and candidate geometric objects, and
where the probability distribution Pr(r.sub.y, s.sub.y|r.sub.x,
s.sub.x) represents the probability that candidate geometric object
y and target geometric x represent cross sections of the same
vessel segment based on scale and response parameter values of the
target and candidate geometric objects. These probability
distributions may be thought of as providing distance,
direction/orientation, and scale/response based tests,
respectively, in the order that they appear in Eqn. 35.
[0179] In some embodiments, the probability distribution Pr(dist(x,
y)|l.sub.x) may be a Gaussian distribution, as shown below, so that
the probability that target geometric object x and candidate
geometric object y represent cross sections of the same vessel
segment decreases exponentially with as the distance between the
target and candidate geometric objects increases. That is, the
probability distribution Pr(dist(x,y)|l.sub.x) may be:
Pr ( dist ( x , y ) l x ) = 1 2 .pi. exp ( ( l x - l y - .mu. ) 2 2
.sigma. d 2 ) . ( 36 ) ##EQU00024##
The probability distribution of Eqn. 36 is a Gaussian probability
distribution having mean .mu. (e.g., 1) and standard deviation
.sigma..sub.d (e.g., 0.3). Though, it should be appreciated that
Pr(dist(x,y)|l.sub.x) may take on any other suitable form and is
not limited to being a Gaussian distribution.
[0180] In some embodiments, the probability distribution Pr({right
arrow over (xy)}|v.sub.x) may follow a super-Gaussian distribution,
which has a "flat" top and exponentially decreasing tails, so that
the probability that target geometric object x and candidate
geometric object y represent cross sections of the same vessel
segment decreases exponentially with increased disparity of
orientation between the target and candidate geometric objects, but
is not sensitive (or at least less sensitive than a Gaussian
distribution) to local variations in the direction/orientation of
centerline voxels (which may occur, for example, due to
digitization errors). That is, the probability distribution
Pr({right arrow over (xy)}|v.sub.x) may be:
Pr ( xy .fwdarw. v x ) = 1 Z exp ( - .theta. ( xy .fwdarw. , v x )
4 .sigma. .theta. 4 ) ) ( 37 ) ##EQU00025##
Though, it should be appreciated that Pr({right arrow over
(xy)}|v.sub.x) may take on any other suitable form and is not
limited to being a super-Gaussian distribution.
[0181] As discussed above, the scale and response parameter values
may be also used to test the viability of linking candidate
geometric object y with target geometric object x. Assuming that
the scale and response values of geometric objects representing
nearby cross sections of the same vessel segment change smoothly,
linking two geometric objects having disparate scale and response
parameter values should be assigned a lower probability. For
example, in some embodiments, the probability distribution
Pr(r.sub.y, s.sub.y|r.sub.x, s.sub.x) may be:
Pr ( r y , s y r x , s x ) = 1 Z exp ( ( s y - s x ) 2 2 .sigma. s
2 ( s ) ) exp ( - ( r y s y 3 - r x s x 3 ) 2 2 .sigma. r 2 ) ( 38
) ##EQU00026##
where Z is the normalization factor and the variance
.sigma..sub.s(s) is equal set, for example, according to max {0.5,
0.2s.sub.x}. Though, it should be appreciated that Pr(r.sub.y,
s.sub.y|r.sub.x, s.sub.x) may take on any other suitable form and
is not limited to having the density function of Eqn. 38.
[0182] Returning to the description of process 2100, after a
candidate geometric object is selected for linking to the target
geometric object, process 2100 proceeds to act 2108 where the
selected geometric object is linked with the target geometric
object. The target and selected geometric objects may be linked in
any suitable way. For example, in some embodiments, the selected
geometric object may be linked with the target geometric object by
storing information identifying the selected geometric object in a
list (or any other suitable data structure) together with
information identifying the target geometric object. Additionally,
in some embodiments, any geometric objects between the selected
geometric object and the target geometric object may be linked to
the target geometric object. For example, in some embodiments, any
geometric object (e.g., Poker Chip.TM.) in a cylinder defined by
the target and selected geometric objects may be linked to the
target geometric object. Any of the geometric objects linked to the
target geometric object may be marked so that when the linking
process continues, the geometric objects already linked and part of
a linked representation of the vessel segment are not considered
again.
[0183] It should be appreciated that after the target and selected
geometric objects (and optionally one or more other geometric
objects between the target and selected geometric objects) are
linked at act 2108, the linked geometric objects form a linked
representation of a vessel segment. In some embodiments, a linked
representation of the vessel segment may be saved for later
processing (e.g., by pushing information identifying the linked
representation into a processing queue). Next, process 2100
proceeds to decision block 2110, where it is determined whether to
continue the linking process by using the selected geometric object
as the target geometric object to further update the linked
representation of the vessel segment to include one or more
additional geometric objects. This decision may be made in any
suitable way, as aspects of the technology described herein are not
limited in this respect. As one example, when the likelihood value
associated with the candidate geometric object (e.g., computed
using Eqn. 33) selected at act 2106 is below a threshold (e.g.,
signifying that none of the candidate geometric objects under
consideration are sufficiently likely to represent a cross section
of the vessel segment), it may be determined to not continue the
linking process by using the selected geometric object as the
target geometric object. On the other hand, when the likelihood
value associated with the candidate geometric object selected at
act 2106 is above a threshold, it may be determined to continue the
linking process by using the selected geometric object as the
target geometric object.
[0184] When it is determined at decision block 2110 that the
linking process is to continue by using the selected geometric
object as the target geometric object, process 2100 proceeds along
the "YES" branch to act 2112, where the target direction is set to
be a direction determined based on the direction/orientation of the
target geometric object and the geometric object selected at act
2106. For example, the target direction may be set to the vector
defined by the location values of the target and selected geometric
objects. For instance, when the target and selected geometric
objects are Poker Chips.TM. having center locations l.sub.x and
l.sub.y, respectively, the target direction may be set to be the
vector v.sub.target=l.sub.y-l.sub.x. Note that each of l.sub.x and
l.sub.y may be a vector indicating the center locations in two or
three dimensions. Next, process 2100 proceeds to act 2114, where
the target geometric object is set to be the geometric object
selected at act 2106. Next process 2100 returns to act 2106 and
acts 2106-2108 and decision block 2110 are repeated. In this way
the linking process may continue so that the representation of a
vessel segment generated by using process 2100 may further be
updated.
[0185] On the other hand, when it is determined at decision block
2110 that the linking process is to not continue by using the
selected geometric object as the target geometric object, process
2100 proceeds to decision block 2116 where it is determined whether
the linking process is to be performed again starting from the
initial geometric object identified at act 2102 but in the opposite
direction from the initial direction identified at act 2102. The
determination to perform the linking process in the "reverse"
direction may be made when the linking process was not previously
run starting with the initial geometric object and an initial
direction that is the opposite direction of the
direction/orientation of the initial geometric object.
[0186] When it is determined that the linking process is to be
continued from the initial geometric object in the opposite
direction from the direction selected at act 2102, process 2100
proceeds to act 2116 where the target geometric object is set to be
the initial geometric object and the target direction is set to be
the opposite direction of the direction/orientation of the initial
geometric object identified at act 2102. Process 2100 then returns
to act 2106. On the other hand, when it is determined that the
linking process is to not be continued from the initial geometric
object in the opposite direction, process 2100 completes.
[0187] The linked representation of a vessel segment obtained by
linking geometry using process 2100 may be used to compute further
geometric features of the vessel segment. For example, the
direction/orientation parameters of geometric objects in the linked
representation capture information about the geometry of the vessel
segment centerline. In some embodiments, integrating the
direction/orientation vectors, a representation of the centerline
curve may be obtained. That is, because the
displacement/orientation vectors may represent tangents to the
centerline curve at each location of a geometric objects in the
linked representation, the centerline curve may be recovered from
linked tangents by integrating over some desired segment of
geometric objects.
[0188] In addition, the linked representation may be used to
determine higher-order and/or more sophisticated geometrical
properties of the vessel segment. For example, derivatives of the
linked orientation vectors may be used to determine the curvature
of the vessel. The centerline curve, length of the curve and
curvature parameters may be used to determine various tortuosity
parameters, which may be used to characterize the vessels.
Moreover, the linked representation of the vessel segment carries
distribution information with respective to the density of vessel
material, the relative distribution of vessels at different radii,
etc. These geometrical, structural and distribution parameters may
be used in a number of ways to analyze vasculature, as discussed in
further detail below. FIG. 13 illustrates a geometrical
representation of vasculature using the linked Poker Chip.TM.
representation, wherein the geometry was extracted from a 3D
volumetric image using the methods described herein.
[0189] Orientation Determination
[0190] As discussed above, linking of centerline voxels (e.g.,
center locations of a Poker Chip.TM.) may be performed according to
criteria that include minimizing the disparity of orientations of
linked centerline voxels. For example, linking of centerline voxels
may be achieved using probability models that provide for a measure
of disparity of direction/orientation between a target Poker Chip
and one or more candidate Poker Chips.TM.. The inventors have
realized that conventional methods for computing
direction/orientation associated with a Poker Chip.TM. (i.e.,
instantaneous vessel direction at the Poker Chip.TM.) may be
unstable and may be unsuitable (or at least inconsistently
applicable) for performing orientation tests for the purposes of
linking (e.g., performed in accordance with process 2100 described
above with reference to FIG. 21).
[0191] Conventional methods of determining orientation are
frequently based on gradient information extracted from the
underlying intensity image. However, computing direction or
orientation from intensity information of an intensity image (e.g.,
a greyscale image) may be unstable and/or inaccurate, particularly
in regions of high curvature and/or high frequency information.
Moreover, direction/orientation computations based on operating on
intensity information are vulnerable to noise. The inventors have
developed a technique for determining direction at a Poker Chip.TM.
from voxel locations, rather than gradient information extracted by
operating on intensity data (e.g., greyscale images).
[0192] According to some embodiments, the direction v of a Poker
Chip.TM. may be computed based on voxel locations in a segmented
image, which herein refers to an image whose voxels are labeled as
corresponding to subject matter of interest or not corresponding to
subject matter of interest (e.g., using zero and non-zero values,
respectively). For example, in images of blood vessels, voxels in a
segmented image may be labeled as corresponding to a vessel or not
corresponding to a vessel. One representation of a segmented image
is a binary image where voxels within a vessel boundary are labeled
as 1 (or a non-zero value, such as a value related to distance from
a vessel boundary, as discussed below) and voxels not within a
vessel boundary are labeled as 0 (or vice versa). When geometry has
been extracted from images (e.g., Poker Chips.TM. have been
obtained from image information), the locations of the voxels
within the cross-section corresponding to the Poker Chip.TM. are
known. This location information may be used to determine
direction/orientation at a Poker Chip.TM. that may be more reliable
than direction/orientation values computed by operating directly on
intensity values in the image(s) (e.g., on the greyscale values of
the image).
[0193] According to some embodiments, displacement vectors are
computed from locations associated with a Poker Chip.TM. by finding
the difference in location between voxel locations associated with
a Poker Chip.TM. and locations in a neighborhood associated with
the Poker Chip.TM.. The neighborhood may be defined in any suitable
way, some examples of which are described in further detail below.
The displacement vectors may be utilized to compute the
direction/orientation of the Poker Chip.TM., for example by
performing principal component analysis (PCA) on the displacement
vectors. However, the displacement vectors may be utilized in other
ways to compute the direction/orientation at the Poker Chip.TM..
Likewise, the location of voxels within a Poker Chip.TM. and a
neighborhood of a Poker Chip.TM. may be used in other ways to
determine direction/orientation at the Poker Chip.TM.. Since the
direction/orientation is computed using location information in a
segmented image(s) and not from intensity information, the
direction/orientation computations may be more accurate and/or
stable (e.g., may be more robust in regions of high curvature, high
frequency, noise and/or other image artifacts).
[0194] The inventors have appreciated that direction/orientation
computations may be made more reliable by operating on a scale
image of the vessel structure. A scale image refers herein to a
segmented image where voxel locations are labeled with a zero
outside the boundary of a vessel (e.g., as determined from
segmentation and/or by extracting vessel geometry from the image)
and are labeled with non-zero values within the vessel boundary
indicating the distance the voxel is from the vessel boundary. For
example, a scale image may be a segmented image for which a
distance transform has been computed such that voxels within a
vessel boundary are labeled with their corresponding distance from
the corresponding vessel boundary. FIG. 27 shows an illustrative
two-dimensional scale image. It should be appreciated that although
the scale image of FIG. 27 is two-dimensional, this is for clarity
of presentation, as scale images may be, and likely are,
three-dimensional images.
[0195] In the scale image of FIG. 27, voxels outside the vessel
boundary are shown in black (e.g., are labeled as zero) and voxels
inside the vessel boundary are labeled with the distance the voxel
is from the vessel boundary. The distance or scale information may
be utilized to define the neighborhood discussed above and/or may
be utilized in evaluating the direction/orientation at a candidate
Poker Chip.TM.. According to some embodiments, displacement vectors
are computed at a candidate location x.sub.0, which may be the
location of a candidate Poker Chip.TM. being assessed for a
possible link to a target Poker Chip.TM.. One exemplary formulation
is provided below.
[0196] According to some embodiments, to determine the
direction/orientation v at a voxel x.sub.0 (e.g., a center voxel of
Poker Chip.TM.) a neighborhood of voxels N(x.sub.0) around voxel
x.sub.0 is computed according to:
N(x.sub.0)={x|distance(x,x.sub.0).ltoreq..left
brkt-top.scale(x)+2.right brkt-bot.} (39)
[0197] As such, a neighborhood of voxels around x.sub.0 may be
defined based on the distance of the voxels from x.sub.0 and the
scale of voxels that are candidates for inclusion in the
neighborhood N(x.sub.0). However, it should be appreciated that a
neighborhood of voxels to around voxel x.sub.0 may be defined in
any other suitable manner, as direction detection techniques are
not limited for use with any particular neighborhood or
neighborhood calculation. From the neighborhood N(x.sub.0) computed
for voxel x.sub.0, a scatter matrix M(x.sub.0) may be computed as
follows:
M ( x 0 ) = i = 1 N ( x 0 ) ( v i - v _ ) ( v i - v _ ) T N ( x 0 )
( 40 ) ##EQU00027##
where v.sub.i=x-x.sub.i for some x.sub.i in N(x.sub.0), v is the
average of the vectors v.sub.i, and .parallel.N(x.sub.0).parallel.
is the number of points in the neighborhood N(x.sub.0). The
direction/orientation v may be associated with the eigenvector
associated with the largest eigenvalue of the scatter matrix M. For
example, the direction/orientation v may be the direction of the
eigenvector associated with the largest eigenvalue (or the opposite
direction). When x.sub.0 corresponds to a center voxel of (or
another voxel representing) a Poker Chip.TM., the direction
orientation v obtained in accordance with the above-described
embodiments may be used as the direction/orientation of the Poker
Chip.TM.. As a result, a direction/orientation v at each candidate
location x.sub.0 may be computed to facilitate linking together of
Poker Chips.TM..
[0198] Branch Point Detection and Linking
[0199] As discussed above, generating a comprehensive linked vessel
structure may involve detecting where vessel segments branch into
further vessel segments and determining how vessel segments are
linked together. The inventors have appreciated that branch point
detection and linking may be achieved using a coarse-to-fine
approach, however, other approaches may be utilized as well.
According to some embodiments, a coarse-to-fine approach includes
detecting branch point candidates from a set of locations (e.g.,
from a set of vessel locations including detected Poker Chips.TM.),
refining the set of branch point candidates based on local behavior
and/or linking branches according to the junction type exhibited
(e.g., the shape/topology of the branch relative to the linked
vessel structure from which it branches). It should be appreciated
that the techniques described herein, though applicable to 2D
datasets, are particularly designed for processing 3D datasets
(e.g., 3D geometry extracted from 3D x-ray scans). Conventional
linking techniques (e.g., those implemented to link roads from a
satellite image) are generally not suitable for accurately linking
vessel structures and are not applicable to 3D datasets.
[0200] FIG. 22 illustrates a method of identifying branch points,
in accordance with some embodiments. In act 2202, a representation
of a centerline for a vessel segment is obtained, the
representation including a plurality of locations (e.g., locations
of geometric objects extracted from a 3D x-ray scan of vasculature)
corresponding to the vessel segment. For example, the
representation of a centerline for a vessel segment may include a
plurality of Poker Chips.TM. linked together using any of the
techniques described herein, wherein the plurality of locations
correspond to the center locations of the respective Poker
Chips.TM.. It should be appreciated, however, that the
representation of a vessel centerline that includes a plurality of
locations may be obtained in other ways, as techniques for
identifying branch locations are not limited for use in connection
with any particular representation or method of obtaining the
representation. The linked segment for which branch points are
being identified is also referred to below as the main segment,
main branch or centerline as needed to clarify from branch segments
or simple curves that may be generated during the course of
identifying branch point for the main linked segment.
[0201] In act 2204, at least one branching score (also referred to
as branch score) is calculated for each of the plurality of
locations of the representation of the vessel segment. A branching
score may be computed in any suitable manner and may depend on the
representation of the vessel segment. According to some
embodiments, the branch score is based, at least in part, on a
measure of asymmetry (e.g., a measure of asymmetric variation in
the distribution of geometric objects (e.g., Poker Chips.TM.)
associated with the representation at each respective location.
According to some embodiments, the branching score is based on the
probability model used to link centerline locations (e.g., Poker
Chips.TM.) as discussed in the foregoing. For example, the
likelihood score from the probability model may be utilized not
only to link centerline locations but to evaluate the likelihood
that locations correspond to a branch point at which a further
vessel segment branches from the representation of the target
vessel segment (main segment) being evaluated. According to some
embodiments, multiple techniques may be utilized to provide a
branching score (e.g., asymmetry and linking likelihood measures
may be used in combination to calculate a branching score or
multiple branching scores). Any suitable technique may be used that
provides at least one branching score indicative of how likely it
is that the associated location corresponds to a branch point or
branch location.
[0202] In act 2206, a set of branch point candidates are identified
based, at least in part, on the branching scores computed in act
2204. For example, branching scores may be evaluated and high
scoring locations (e.g., via thresholding or by taking the N
largest branch scores) may be identified as branch point
candidates. According to some embodiments, local maxima of the
computed branching scores are identified to select the set of
branch point candidates. For example, the branching scores computed
for the plurality of locations may be viewed as a function from
which branch point candidates may be selected by identifying local
maxima (e.g., peaks) in the function. In embodiments in which
multiple branching scores are computed for each location, the
branching scores may be combined (e.g., via a weighted sum, average
and/or weighted average) to form a single branching score for each
location from which branch point candidates may be selected (e.g.,
using thresholding, N-greatest scores, local maxima techniques,
etc.). Alternatively, multiple branching scores at each location
may be analyzed in other ways, such as using a rule-based approach
that considers the values of each branching point at a location and
determines whether the location should be identified as a branch
point candidate, using fuzzy logic, or any other suitable technique
for evaluating multiple branch scores so as to determine whether a
location should be considered (at least preliminarily) as a branch
point candidate.
[0203] According to some embodiments, acts 2204 and 2206 may
reflect a coarse process to identify a set of branch point
candidates, which set may be further refined using further
processes (e.g., by performing one or more of acts 2208 and 2210
described in further detail below) to eliminate one or more of the
branch point candidates that do not meet further (and perhaps
stricter) criteria. According to some embodiments, the set of
branch point candidates computed, for example, in the manner
described in connection with act 2204 and 2206 may be used as
representing the final set of branch points, as further refinement
may not be necessary in certain situations or environments.
However, in some circumstances, it may be desirable to refine or
filter the branch point candidates to remove from further
consideration branch point candidates that do not exhibit one or
more further characteristics of a branch point and/or to correctly
characterize the branch segments with respect to the branch
segments relationship with the centerline, some examples of which
are described in further detail below.
[0204] In act 2208, one or more characteristics of the identified
branch point candidates may be analyzed to refine the set of branch
points to include only those branch point candidates that meet
further criteria. According to some embodiments, the behavior of
surrounding locations (e.g., surrounding Poker Chips.TM.) may be
evaluated to assess whether the branch point candidate is more
likely to represent a true branch point, a non-branch point, or a
nearby branch point. For example, each branch point candidate may
undergo further processing by locally linking unlinked geometric
objects in the vicinity of the branch point candidate (e.g., using
any of the techniques described above in connection with FIG. 21)
to evaluate whether the locally linked vessel segment (e.g., simple
curve) interacts with the main linked vessel segment (centerline)
from which it has been identified as a branch candidate in a manner
suggestive of a branch (e.g., whether the linked branch segment, or
a projected trajectory of the linked branch segment, intersects
with the existing linked vessel segment) and/or how the locally
linked vessel segment(s) (e.g., minor branch(es)) interact with the
main vessel segment (e.g., major branch). Other characteristics of
identified branch point candidates may be additionally or
alternatively considered (e.g., location, orientation and/or scale
of proximate geometric objects may be evaluated for continuity with
an existing linked vessel segment) to refine the set of branch
point candidates, as the act of refining is not limited to
consideration of any particular characteristic or set of
characteristics associated with the branch point candidates.
[0205] Refinement of the branch point candidates may include
determining a type of junction for the branch point. For example,
vessel structures may branch in a number of ways, each exhibiting a
different behavior at the junction (branch point). Identifying the
junction type may be performed to facilitate correctly linking
vessel segments at their respective branch points. In this manner,
branch point detection may be performed using coarse processing
followed by a refining process to implement a coarse-to-fine
approach, some examples of which are described in further detail
below.
[0206] In act 2210, the refined set of branch point candidates may
be further processed to remove outliers, though act 2210 may not
need to be performed. In particular, the identified branch points
and accompanying branch segments (e.g., simple curves) may be
further analyzed to remove outliers from consideration as valid
branch points or to identify an actual branch point when two or
more branch points correspond to a same simple curve. For example,
simple curves identified as branching from a branch point on a
target centerline may be removed as potential branches when it is
determined that insufficient non-zero scale voxels exist between
the first geometric object in the simple curve and the branch point
to support the hypothesis that the simple curve is a branch of the
target centerline. This situation of insufficient support is
indicative of the simple curve belonging to a distinct vessel
segment other than the target linked segment (current
centerline).
[0207] According to some embodiments, identified branch points are
further evaluated to assess whether branch points that are close
together are true branch points or whether they correspond to a
single branch points. For example, adjacent branch points along a
centerline that meet a proximity requirement are further evaluated
to identify the true branch point. In other embodiments, multiple
simple curves that have been associated with a single branch point
are evaluated to assess the true branching structure at the branch
point. Any of various methods may be utilized to identify outliers,
filter identified branch points or otherwise rectify the branch
points and associated minor branches to accurately reflect the true
branching structure. Method 2200 in FIG. 22 described above may be
repeated on any number of linked vessel segments so as to obtain a
comprehensively linked structure of geometric objects (e.g., Poker
Chips.TM.) extracted from, e.g., a 3D x-ray scan of vasculature
and/or otherwise obtained.
[0208] As discussed above, identifying branch points may include a
coarse-to-fine approach wherein a coarse process is performed to
identify branch point candidates and a refining process is
performed to identify branch points based on the branch point
candidates. As also discussed above, the coarse process may include
identifying branch point candidates using a branch point scoring
process. According to some embodiments, the branch point scoring
process is based on the inventors' recognition that the
distribution of geometric objects (e.g., Poker Chips.TM.) at branch
locations will generally exhibit different characteristics than
geometric objects (e.g., Poker Chips.TM.) along a vessel segment at
non-branch locations. Differences in Poker Chips.TM. distribution
may be captured in a number of different ways including examining
how Poker Chips.TM. are distributed about a target Poker Chip.TM.
along a centerline of a vessel segment. FIG. 23 illustrates a
method of identifying branch point candidates, in accordance with
some embodiments. Method 2300 may be performed, for example, on a
set of geometric objects (e.g., Poker Chips.TM.) extracted from a
3D x-ray dataset of a vasculature network, or may be performed on
geometric objects obtained from other datasets.
[0209] In act 2310, a target geometric object is selected for
analysis. The target geometric object may be one of the centerline
locations of a linked vessel segment, linked in accordance with any
of the techniques described herein. For example, each centerline
location in a linked vessel (e.g., represented by a linked segment
of Poker Chips.TM.) may be evaluated for possible branching at that
location and the selected target geometric object may be a first
geometric object in the linked segment to be evaluated. The
evaluation may then be performed iteratively on each geometric
object in the linked segment or on a desired number of geometric
objects in the linked segment (main segment or centerline).
[0210] In act 2320, the distribution of geometric objects in a
neighborhood of the target geometric is evaluated to determine at
least one characteristic of the distribution. As discussed above,
the inventors have recognized that geometric objects are
distributed differently at branch locations than at non-branch
locations. Based on this insight, the inventors have developed
techniques to determine one or more characteristics of the
neighborhood distribution of geometric objects to score the
location associated with the target geometric object according to
how strongly the neighborhood distribution is suggestive of a
branch point. FIG. 24A illustrates schematically the evaluation of
geometric objects in a neighborhood of a target geometric
object.
[0211] The neighborhood can be chosen to be of any size, but
typically is chosen to be large enough so as to accurately capture
the local distribution of geometric objects. According to some
embodiments, the neighborhood is selected to be of a fixed size
about the target geometric object (e.g., all geometric objects
within a circle centered on the target geometric object and having
a fixed radius). According to some embodiments, the neighborhood is
selected to be of size that depends, at least in part, on one or
more of the parameters that describe the geometric object. For
example, when the geometric object corresponds to a Poker Chip.TM.,
the neighborhood size may be selected based at least in part on the
scale (e.g., radius) of the Poker Chip.TM.. However, the size of
the neighborhood may be selected in any manner suitable for a given
application.
[0212] According to some embodiments, displacement vectors between
the target geometric object and geometric objects in a
predetermined neighborhood of the target are evaluated to determine
one or more characteristics of the distribution. A displacement
vector refers herein to any vector that provides information
regarding a spatial relationship between locations. As such, in the
context of geometric objects, a displacement vector may be any
vector construct that provides information regarding the spatial
relationship between geometric objects. For example, a displacement
vector may capture the spatial relationship between two Poker
Chips.TM. (e.g., the spatial relationship between the center
locations of two Poker Chips.TM.). The inventors have appreciated
that displacement vectors between a target geometric object and
geometric objects in a neighborhood of the target provides a
mechanism for capturing the manner in which the geometric objects
are distributed about the target geometric object, and may be used
as a basis for computing one or more characteristics of the local
distribution of geometric objects.
[0213] Typically, the displacement vectors will more generally
align in the direction of the centerline (main segment) of a linked
structure at non-branching points and will have more variability in
this respect at branch points. Accordingly, this property may be
exploited to score the location associated with the target
geometric object based on the variability (also referred to herein
as asymmetry) in connection with the distribution of geometric
objects in the neighborhood. According to some embodiments, the
variation in the principal directions of the displacement vectors
is computed to obtain a measure of the variability or asymmetry of
the neighborhood about the target geometric object. Any technique
may be used to evaluate the variability of a neighborhood of
geometric objects, some examples of which are discussed in further
detail below.
[0214] In act 2330, the location corresponding to the target
geometric object is scored relative to how indicative the at least
one property of the target neighborhood is of a branch point.
According to some embodiments, the variability (or asymmetry) of
the neighborhood of geometric objects about and in relation to the
target geometric object, however computed, is used to score the
location associated with the target geometric object. The score may
then be utilized to identify a set of branch point candidates, as
discussed in further detail below.
[0215] In act 2335, it is determined whether there are further
target geometric objects to be evaluated. For example, it may be
determined whether each geometric object that forms a linked
segment has been considered in connection with branch scoring
(e.g., whether there are further Poker Chips.TM. along a linked
vessel segment for which branch scoring has not been performed). If
there are further geometric objects to consider, a next target
geometric object is selected for scoring (act 2310) and a
neighborhood of the new target geometric object is evaluated (act
2320) and a location associated with the new target geometric
object is branch scored (act 2330). If all of the desired geometric
objects (e.g., all Poker Chips.TM. along a linked segment) have
been branched scored, the resulting branch scores may be further
evaluated to identify a set of branch point candidates. FIG. 24B
illustrates a number of geometric objects and FIG. 24C illustrates
the geometric objects shaded in relationship to their respective
branch scores using some embodiments of branch scoring.
[0216] In act 2340, the branch scores computed by iteratively
repeating acts 2310-2330 (e.g., branch scoring each Poker Chip.TM.
along a linked vessel segment) may be evaluated to locate branch
point candidates. For example, all locations having a branch score
that meets certain criteria may be selected as branch point
candidates. The criteria selected may be any criteria that suitably
identifies locations that are indicative of branch points.
According to some embodiments, the branch scores are thresholded
and all locations that meet the threshold criteria (e.g., that
exceed a given threshold) are selected as branch point candidates.
According to some embodiments, the branch scores are viewed as a
function and local maximum (or local minimum) are identified and
selected as branch point candidates (e.g., peak picking may be
performed on the computed branch scores). However, any technique
that suitably identifies branch point candidates from the scored
locations may be used, as identifying branch point candidates from
a plurality of branch scores is not limited to any particular
technique(s) for doing so.
[0217] As discussed above, the inventors have appreciated that
candidates for possible branch points (i.e., points in a vessel
structure where a vessel branches into two or more vessels,
including splits, tributaries, multiple splits, etc.) may be
determined in a number of ways. According to some embodiments,
branch point candidates are identified at locations where the
linking analysis described above does not result in any high enough
probability link to another Poker Chip.TM. (e.g., according to some
threshold of likelihood). As such, branch point candidates may
include the termination points of linked segments.
[0218] The inventors have also appreciated that branch points may
also exhibit different variation and/or asymmetry properties than
non-branch locations (e.g., with respect to the distribution of
geometric object about such points). In view of this insight, the
inventors have developed techniques to evaluate variation patterns
at locations along a linked segment to facilitate identifying
branch point candidates. As also discussed above, the inventors
have recognized that the distribution of geometric objects differs
at branch locations than at non-branch locations (or may exhibit
other asymmetry characteristics that are detectable). For example,
the principal directions of variation for displacement vectors
computed in connection with a neighborhood of a target geometric
object may be computed to assess how the geometric objects in the
neighborhood are distributed. According to some embodiments, the
principal directions of variation for displacement vectors computed
between a target geometric object and each geometric object in a
neighborhood of the target geometric object is computed and
analyzed to assess whether the target geometric object corresponds
to a possible branch point or branch location. Following below are
non-limiting examples of analyzing a neighborhood of geometric
objects about a target location to assess whether the distribution
of the geometric objects indicates the presence of a branch point.
According to some embodiments, principal component analysis may be
performed on displacement vectors in a neighborhood of a target
geometric object being considered to evaluate geometric object
distribution and/or structure about the target geometric object,
for example, by identifying asymmetry based on the relationship of
the principal components (e.g., based on principal directions of
variation as assessed by comparing eigenvectors and/or eigenvalues
of a matrix formed from a neighborhood of the target geometric
object). According to some embodiments a segmented image and/or a
scale image may be used to define a neighborhood and/or compute
displacement vectors from which the principal directions of
variation may be determined. The principal directions of variation
may be evaluated (e.g., using eigenvector analysis) to assess
whether the target geometric object corresponds to a possible
branch point location.
[0219] According to some embodiments, detection of branch point
candidates may proceed by defining a neighborhood of a target voxel
x.sub.0 associated with a geometric object (e.g., centered on a
Poker Chip.TM.) as follows:
(x.sub.0)={x|.sigma.(x.sub.0)<dist(x,x.sub.0).ltoreq.Dconnect(x,x.sub-
.0)} (41)
[0220] Where the distance D may be expressed as follows:
D = { 2 .sigma. ( x 0 ) + 4 .sigma. ( x 0 ) < 4 .sigma. ( x 0 )
+ 8 .sigma. ( x 0 ) .gtoreq. 4 ( 42 ) ##EQU00028##
[0221] That is, the neighborhood of a target voxel x.sub.0 centered
on a Poker Chip.TM. may be defined by non-zero scale voxels having
a distance from target voxel x.sub.0 between .sigma.(x.sub.0) and
D. It should be appreciated that a neighborhood of a target voxel
may be computed in any way and the above formulation is merely one
example of defining a neighborhood N. Typically, voxels in the
neighborhood of the target voxel (i.e., the location associated
with the target geometric object) that are considered in the
following computations are those that are also associated with the
location of a geometric object (e.g., voxels that represent the
center location of a Poker Chip.TM.). For each x in the
neighborhood N (however computed) a displacement vector may be
computed as follows:
v = x - x 0 x - x 0 ( 43 ) ##EQU00029##
[0222] As discussed above, to assess characteristics of the
geometric objects in a neighborhood, displacement vectors may be
computed only for voxels associated with a geometric object.
However, in other embodiments, displacement vectors may be computed
for each voxel in the neighborhood or for some other desired subset
of voxels in the neighborhood. The displacement vectors may be used
to form a matrix suitable for performing principal component
analysis. According to some embodiments, the matrix may be
formulated as follows:
M ( x 0 ) = i .di-elect cons. ( x 0 ) vv T ( x 0 ) ( 44 )
##EQU00030##
[0223] Where .parallel.N(x.sub.0).parallel. is the number of voxels
inside the neighborhood for which a displacement vector is
computed. It should be appreciated that the above matrix is one
example of a matrix that may be suitable for providing a basis for
performing principal component analysis to evaluate characteristics
of neighborhood N. When M is computed as a real symmetric matrix as
it is in the above formulation (48), the matrix can be diagonalized
as:
M = U ( .lamda. 1 .lamda. 2 .lamda. 3 ) U T ( 45 ) ##EQU00031##
[0224] The eigenvalues .lamda. and/or the relationship between the
eigenvalues may be analyzed to assess one or more properties
regarding the principal directions of variation of the displacement
vectors, e.g., to assess whether the vessel structure at the target
location is symmetric with respect to its cross-section (likely no
branch point) or asymmetric in this respect (branch point
candidate). For example, when x.sub.0 is not on or not close to a
branch point, .lamda..sub.2 and .lamda..sub.3 will likely be small
(e.g., close to zero). When x.sub.0 is near or at a branch point,
.lamda..sub.2 will likely be relatively large. As such,
.lamda..sub.2 may operate as an indicator as to the likelihood of
the existence of a branch point at location x.sub.0. According to
some embodiments, the principal components are computed for visited
Poker Chips.TM. such that the values .lamda..sub.2 form a function.
The peaks of this function may be selected as branch point
candidates for further evaluation (e.g., fed into a branch point
model to assess whether the locations correspond to branch points
and/or to assess what type of branch point the candidate
represents, as discussed in further detail below. A sliding window
may be utilized to evaluate branch scores (e.g., branch scores
based, at least in part, on .lamda..sub.2) determined for each
desired target voxel x.sub.0 (e.g., visited Poker Chips.TM.) and
the local maximums in the resulting branch score function (e.g.,
asymmetry function) selected as branch point candidates. It should
be appreciated that the above described technique for evaluating
the variation with respect to the distribution of geometric objects
(e.g., the principal directions of variation of displacement
vectors) in a neighborhood of a target geometric object is a
non-limiting example and other ways of evaluating characteristics
of the distribution of geometric object may also be performed to
facilitate identification of branch point candidates.
[0225] As discussed in the foregoing, identifying branch point
candidates (e.g., as described in connection with FIG. 23 above)
may represent a coarse process that identifies branch point
candidates for further consideration, or may represent a final set
of branch points without undergoing further refinement. According
to some embodiments, the set of branch point candidates identified
via branch scoring undergo further refinement to evaluate whether
the branch candidate is a branch point and how an accompanying
minor branch interacts with the presently linked vessel at the
branch point. FIG. 25 illustrates a method 2500 of refining a set
of branch point candidates and identifying relationships between
branch junctions and a linked segment, in accordance with some
embodiments. According to some embodiments, a branch model is
utilized to identify different junction types, as also discussed in
further detail below.
[0226] In act 2502, a branch point candidate is selected for
evaluation. For example, the branch point candidate may be selected
from a set of branch point candidates identified according to any
of the techniques described herein (e.g., the branch point
identification method described in connection with FIG. 23). In act
2504, unlinked geometric objects in a neighborhood of the branch
point candidate are identified. The neighborhood may be the same or
different than the neighborhood selected for branch scoring and
identification of branch point candidates. In act 2506, the
identified unlinked geometric objects in the neighborhood are group
into simple curves. The term "simple curve" refers herein to
locally linking geometric objects in the neighborhood. For example,
any of the techniques described herein for linking geometric
objects (e.g., linking Poker Chips.TM.) may be applied to form
relatively short segments from the unlinked geometric objects in
the neighborhood.
[0227] According to some embodiments, the unlinked geometric object
in the neighborhood that is closest to the branch point candidate
is selected to begin local linking, after which the next closest
unlinked geometric object may be selected for local linking and so
on until all of the unlinked geometric objects in the neighborhood
have been locally linked. According to some embodiments, the
unlinked geometric object with the largest radius (or closest
radius to the geometric object at the branch point candidate), the
highest confidence, or any other criteria may be selected to
perform local linking and repeated until each unlinked geometric
object in the neighborhood has been processed to form one or more
linked segments (simple curves). Linking of each individual simple
curve from a starting point (e.g., the closest unlinked geometric
object from the linked vessel segment (centerline) may be
terminated using any type of criteria. For example, local linking
of a simple curve may be stopped when the linking probability falls
below a certain threshold, using a distance threshold (e.g.,
distance from the centerline), using a maximum length criteria for
the simple curve, using a maximum number of geometric objects
criteria, etc. The simple curves (however computed and in whatever
order linked) may then be assessed to evaluate their interaction
with the linked vessel segment to assist in identifying the branch
points and/or the type of branch corresponding to the simple
curve.
[0228] In act 2508, the geometric object in each simple curve
closest to a geometric object in the linked vessel segment is
identified. That is, for each simple curve generated in act 2506,
the pair of geometric objects on the simple curve and the linked
vessel segment having a minimum distance are identified. This
information may be utilized to identify the actual branch points
and also to characterize the type of junction formed (e.g., using a
junction-type or branch model). In particular, the geometric object
of a linked structure closest to a geometric structure on a simple
curve may be identified as a branch point. The geometric object
identified as a branch point may correspond to a branch point
candidate or may correspond to another geometric object on the
linked segment that was not identified as a branch point candidate.
With the branch points on the linked segment identified, the branch
or junction-types may be identified to correctly link together the
larger vessel structure. According to some embodiments, the simple
curves (e.g., relatively short locally linked segments (minor
branch), grown from unlinked geometric objects in the neighborhood
of a branch point candidate) may be projected or extended along its
trajectory to determine whether the simple curve intersects with
the linked vessel structure to assess whether the simple curve is a
branch of the linked vessel structure.
[0229] According to some embodiments, determining whether and where
a simple curve branches from a linked vessel segment or centerline
of a major branch proceeds by considering a ray R, for example,
between the closest points. Given a simple curve or line segment of
a possible minor branch (a set of points x.sub.i, i=1, . . . , n)
and a location on a linked vessel segment (centerline) x.sub.0
(e.g., a location associated with a geometric object on a linked
vessel segment/major branch), the best ray fitted to points set
{x.sub.i} and passing x.sub.0 is obtained by solving the
minimization problem.
arg min dist ( x i , R ( x 0 , v ) ) = arg min v { ( x i - x 0 ) 2
- [ ( x i - x 0 ) v ] 2 } = arg max v [ ( x i - x 0 ) v ] 2 = arg
max v ( v T [ ( x i - x 0 ) ( x i - x 0 ) T ] M v )
##EQU00032##
[0230] The solution to above minimization is the eigenvector
associated with the largest eigenvalue. A measure of the likelihood
that the simple curve (minor branch) actually joins (i.e., branches
from) the linked vessel segment may be defined as follows.
1 2 .pi. exp ( i d i 2 2 ) ##EQU00033##
[0231] where d.sub.i is the residual error of each point on the
possible branch segment to the best fitted ray R. Accordingly, each
geometric object location on the centerline L, may be evaluated to
find the location with the smallest fitting error
e = d i 2 N - 1 . ##EQU00034##
If all of the fitting errors
e > 2 2 , ##EQU00035##
there is no branching point. Otherwise, the a set of points, S, on
the centerline L which have the fitting error falling in the range
[e,e+0.1] are identified (e.g., all geometric objects on the linked
vessel structure that have an error below a threshold are
identified as possible branch points). When multiple possible
branch points are identified, further processing may be performed
to identify one or more actual branch points and/or to characterize
the type of junction at the identified branch point, as discussed
in further detail below.
[0232] In act 2510, the type(s) of the identified branch points are
determined. For example, the junction at which a vessel segment
branches from another segment may take on a number of different
configurations. FIGS. 26A-C illustrate examples of junction types
of a branching vessel structure. FIG. 26A illustrates a Y-junction,
FIG. 26B illustrates a V-junction and FIG. 26C illustrates a
T-junction. The inventors have developed techniques for identifying
these junction types based on characteristics of the identified
branch points and the simple curves to facilitate accurately
constructing a linked vessel structure. In some embodiments, a best
matching vessel segment model may be fitted to the branch points in
a neighborhood of an identified branch point.
[0233] According to some embodiments, a branch model or junction
model is utilized to classify the branch type of each identified
branch points, as discussed in further detail below. For example,
given a neighborhood of a branch point candidate and a centerline L
(e.g., a linked vessel segment), a local linking algorithm is
applied to all unlinked geometric objects in a neighborhood to
generate linked segments {c.sub.i, i=1 . . . n} (e.g., acts
2502-2506 may be performed). For each segment c.sub.i, the location
of the geometric object in the centerline or major linked vessel
segment (e.g., x.epsilon.L) that has the minimum distance to the
curve c.sub.i is found (e.g., act 2508 may be performed). If x is
in the middle of the linked vessel segment L, the junction between
the linked vessel segment L and the simple curve c.sub.i is a
Y-junction (e.g., as shown in FIG. 26A). If x is at the end of the
linked vessel segment L, the junction between the linked vessel
segment L and the simple curve c.sub.i is a V-junction (e.g., as
shown in FIG. 26B). Otherwise, the junction between the linked
vessel segment L and the simple curve c.sub.i is a T-junction
(e.g., as shown in FIG. 26C).
[0234] The inventors have developed a branch point model to
evaluate the above circumstances and also to handle certain special
cases that may arise. For example, the following procedure may be
utilized to identify junction types in a variety of circumstances.
As discussed above, the pair of locations on the linked vessel
segment (centerline) and a simple curve (possible minor branch)
that have a minimum distance may be identified. The linked vessel
structure and the simple curve are referred to in the following
description as the centerline (or major segment) and the simple
curve (or minor segment), and the two locations are referred to as
the major closest point and the minor closes point, respectively.
When the major and minor closest points are in the middle of the
centerline and the simple curve, respectively, the simple curve
centered on the minor closest point may be broken and only the part
considered a straight line is taken for further evaluation. This
straight line segment may be evaluated as a simple Y junction by
the Y-junction model. When the major closest point is at one end of
the centerline, and the minor closest point is in the middle of the
minor simple curve, the branch point may be labeled as a T-junction
type branching point. When both major closest point and minor
closest point are at one end of the centerline and minor simple
curve, respectively, the branching point may be labeled as a
T-junction type branch point. It should be appreciated that the
circumstances that arise regarding the structure and configuration
of branch points may be evaluated and resolved in other ways and
the foregoing description merely describes some possible ways of
doing so.
[0235] Various techniques described in the foregoing may be
utilized to obtain a comprehensively linked structure, e.g., a
fully linked vessel network for a 3D x-ray scan of vasculature.
However, the techniques may also be used to linked together only
portion of a vasculature, as the techniques described herein on not
limited for use to any particular linking application or
result.
[0236] Detection of Loop Structure
[0237] As discussed above, a linked representation of a vessel
network may comprise a network structure graph representing
connectivity among vessel segments in the vessel network. The
inventors have recognized that loops in a vessel network may be
difficult to detect and conventional linking methods were not
equipped to detect or handle such loop structures in the vessel
network. As a result, conventional methods of linking vessel
structures are inaccurate in this respect.
[0238] Accordingly, in some embodiments, loops in a vessel network
may be detected and taken into account when generating a linked
representation of the vessel network. In this way, the vessel
network structure graph may accurately represent loops in the
vessel network (e.g., the vessel network structure graph may
comprise one or more cycles) representing loops in the vessel
network).
[0239] Loops in the vessel network may be detected in any of
numerous ways. For example, according to some embodiments, loops
may be detected in part by labeling geometric objects (e.g., Poker
Chip.TM.) as visited and/or linked such that when a geometric
object that is already labeled as visited and/or linked is
identified as a candidate geometric object to be linked to more
than a single vessel segment (e.g., two or more different vessel
segments), the geometric object may be evaluated from both
directions (that is with respect to membership to each of two or
more vessel segments) to assess whether the vessel structure forms
a loop. When it is determined that a geometric object may be linked
to two or more vessel segments (which may be done in any suitable
way including, for example, the linking techniques described with
reference to FIG. 21 above), the location of the geometric object
(e.g., the center point of a Poker Chip.TM.) may be identified as a
branch point location. In such a case, the vessel network structure
graph may be updated to include a vertex corresponding to the
branch point and to include edges, incident to this vertex, that
correspond to the two or more vessel segments to which the
geometric object may be linked to.
[0240] Parallelization
[0241] According to some embodiments, the linking algorithm may be
performed in parallel. Since linking is generally local and may not
need to rely on the information from far away voxels, the algorithm
can be parallelized by dividing the image into small blocks. Then
individual CPUs may operate on a single block without the need to
communicate with other blocks. Because of the computation requires
some neighborhood information, each block may include a fixed
margin overlapping with its neighbor's margin. The speed gained by
parallelization is the number of processors divided by one plus
overhead caused by margin. In one example, dividing a volume of
2000.times.2000.times.1400 into 500.times.500.times.500 blocks and
using 8 processors produced a gain of 4.49 times processing
speed.
[0242] The margin for parallelization may be chosen based on the
following: 1) the margin for the scale selection
m.sub.s=r.sub.max+1; 2) the margin for the smoothing
m.sub.sm=3.sigma.; 3) the margin for the gradient computation mg=1;
4) the margin for the direction detection
m.sub.d=m.sub.g+r.sub.max+1+m.sub.sm; 5) the margin for centerline
filtering m.sub.c=max {2r.sub.max, m.sub.d}; and 6) the margin for
the non-maximum suppression m.sub.sprs=r.sub.max+m.sub.c.
[0243] Because the block algorithm for parallelization needs to
divide the volume into blocks at the beginning and assembling the
blocks into a volume at the end, a way to transform between global
coordinates and block coordinates may be needed. The block id
(b.sub.x, b.sub.y, b.sub.z) for a point (i, j, k) in the global
coordinate is given as:
b x = i s b y = j s b z = k s ( 46 ) ##EQU00036##
The local coordinates in its block is (i', j', k')
i'=i-b.sub.xs
j'=j-b.sub.ys
k'=k-b.sub.zs (47)
[0244] The dimension (s.sub.x, s.sub.y, s.sub.z) of the block
(b.sub.x, b.sub.y, b.sub.x) is:
s x ( b x ) = { mod ( N x , s ) if b x = N x s - 1 N x s .noteq. 0
0 if b x < 0 s otherwise s y ( b y ) = { mod ( N y , s ) if b y
= N y s - 1 N y s .noteq. 0 0 if b y < 0 s otherwise s z ( b z )
= { mod ( N y , s ) if b z = N z s - 1 N x s .noteq. 0 0 if b z
< 0 s otherwise ( 48 ) ##EQU00037##
[0245] Given a point (i', j', k') at block (b.sub.x, b.sub.y,
b.sub.z), the global offset in the file is:
pos = i ' s y s z + j ' s z + k ' + ( b z N x N y s z ( b z - 1 ) +
b y N x s y ( b y - 1 ) s z ( b z ) + b x s x ( b x - 1 ) s y ( b y
) s z ( b z ) block offset ) ( 49 ) ##EQU00038##
[0246] The number of blocks in the x dimension is
n bx = N x s , ##EQU00039##
the number of blocks in the y dimension is
n by = N y s ##EQU00040##
and the number of blocks in the z dimension is
n bz = N z s . ##EQU00041##
A one dimensional block ID 1=(1, . . . , n.sub.bxn.sub.byn.sub.bz)
to 3D index
b x = l n by n bz b y = l - b x n by n bz n bx b z = l - b y n bz -
b x n by n bz ( 50 ) ##EQU00042##
[0247] Three dimensional block ID (b.sub.x, b.sub.y, b.sub.z) to
one dimensional block ID. When connecting the linked structure from
adjacent blocks, a glue layer between the different blocks may be
analyzed and the how the linked structure in the adjacent blocks
approach Poker Chips.TM. in the glue layer may be evaluated to
determine how to link the local linked structures together.
According to some embodiments, probability models (e.g., similar to
the probability models discussed above) can be used to assess the
likelihood that a Poker Chip.TM. in the glue layer is part of local
linked structure in two or more adjacent blocks.
[0248] The inventors have appreciated that the speed of linking a
geometric representation of a vessel structure may be accelerated
by dividing the representation into smaller regions and processing
them in parallel. The inventors have developed techniques for
stitching the linked structures from the smaller regions together
to form a larger linked structure representing the vessel network.
Methods for stitching or gluing structures from adjacent regions
together are described in further detail in Appendix A. According
to some embodiments, location and direction of Poker Chips.TM. in a
glue region at the juncture of adjacent regions are evaluated to
determine how sub-structures should be stitched or glued together
to form a larger linked structure.
[0249] Information relating to the geometry of a subject's
vasculature, or a portion thereof, can be used to determine one or
more qualitative and/or quantitative measures of geometrical,
structural, and/or distribution parameters of the subject's
vasculature that are informative for diagnostic, predictive,
prognostic, therapeutic, interventional, research and/or
development purposes, as well as for grading and/or staging a
disease. It should be appreciated that vasculature geometry may be
obtained for any suitable blood vessel volume, as the aspects of
the technology described herein are not limited in this respect. In
some embodiments, all the geometrical information captured by the
linked Poker Chips within a target volume of interest may be
evaluated. However, in some embodiments, useful information may be
obtained from analyzing only a subset of Poker Chips within a
target volume (e.g., about 10%, about 20%, about 30%, about 40%,
about 50%, about 60%, about 70%, about 80%, or about 90%) as
aspects of the technology described herein are not limited in this
respect.
[0250] According to aspects of the technology described herein, the
types of geometrical or structural information that may be
extracted from images (e.g., extracted from a linked Poker Chip
representation) includes a measure of vessel curvature, tortuosity,
branching, diameter, etc., or any combination thereof. Optionally,
or additionally, a measure of vessel density (and/or the density of
vessels having one or more predetermined structural
characteristics) may be determined and/or analyzed. It should be
appreciated that a Poker Chip may consist of or include information
relating to the size (radius), angle, etc. of the vessels being
represented. In some embodiments, the Poker Chip representation may
include linking information (e.g., relating to the linkage angle
etc. between a first Poker Chip and one or more adjacent Poker
Chips).
[0251] Tubular structures (e.g., blood vessels in a cast or in
vivo) of different size ranges may be analyzed separately and
compared to different threshold or reference values as described
herein. In some embodiments, one or more structural parameters are
obtained (e.g., calculated or modeled, etc.) for only a subset of
size ranges (e.g., only for those size ranges for which changes are
known to be associated with a diagnostic, prognostic, clinical, or
research application of interest). However, in certain embodiments,
all of the size ranges are analyzed. In some embodiments, one or
more different parameters are analyzed for different size ranges.
However, in certain embodiments, the same parameter(s) is/are
analyzed for all of the size ranges that are being assayed.
Analyses may be provided in the form of histograms or curves
representing a distribution of numerical values or scores obtained
for the different ranges.
[0252] It should be appreciated that analytical techniques used to
categorize blood vessels based on size may be used to categorize
other tubular body structures based on size. In some embodiments,
once the tubular structures (e.g., blood vessels) are categorized
based on size, the associated values or scores obtained for
different parameters of interest can also be categorized and
analyzed. Aspects of the technology described herein may be
automated, for example, as described herein.
[0253] Aspects of the technology described herein relate to
analyzing data obtained for body structures in animals (e.g., in
test animals). In one embodiment, the technology described herein
relates to obtaining pattern information relating to one or more
aspects or regions of the vasculature of an animal. Pattern
information obtained according to aspects of the technology
described herein may be used to analyze a disease model (e.g., to
assess whether an animal disease model is representative of an
actual disease based on structural vascular features, or to assess
the progression of one or more vascular changes in a test animal
that provides a validated disease model, etc.), to evaluate the
effectiveness of a treatment regimen, to identify candidate
compounds or treatment regimens that are therapeutically effective,
or for other applications where data relating to vascular
structures (e.g., the progression of vascular structures, changes
in vascular structure over time or in response to different drugs
or drug dosages or administration frequencies, etc., or any
combination thereof) is informative. For example, aspects of the
technology described herein may be used to identify one or more
pattern elements that can be used to help diagnose or evaluate
diseases, including but not limited to cancer, retinopathies, and
cardiac, renal, and/or haptic disease, provide prognostic
information, monitor treatments, screen therapeutic agents, select
one or more therapeutic agents (e.g., help determine or predict a
subject's responsiveness to a particular drug), etc., or any
combination thereof.
[0254] In some embodiments, structural vascular features, and/or
changes in structural vascular features, can be used to evaluate
the effectiveness and/or toxicity of one or more therapeutic
compounds or treatment modalities. In some embodiments, the
toxicity of a compound (e.g., a known therapeutic compound or a
candidate therapeutic compound) can be evaluated by determining
vascular changes in response to the compound. The vascular changes
can be determined over the whole body, within a tissue, within an
organ (e.g., the liver or kidneys), or within a portion of any one
thereof. In some embodiments, a qualitative assessment of vascular
change is made. In some embodiments, a quantitative assessment of
vascular change is made. In some embodiments, vascular changes in a
healthy body, tissue, or organ, is evaluated. In some embodiments,
toxicity (e.g., drug toxicity) can be determined based on changes
in vascular patterns (e.g., changes in vascular morphology or any
other change in vascular features described herein). In some
embodiments, a vascular therapeutic index can be calculated as a
ratio between vascular changes in treated diseased regions versus
vascular changes in normal, non-treated, tissues, organs or organ
regions. In some embodiments, a ratio of vascular changes in a
treated diseased region (e.g., a tumor) relative to vascular
changes in a control (e.g., either a control that is not treated,
or a control tissue that is not diseased but that is exposed to the
treatment) can be calculated. In some embodiments, vascular changes
in a non-diseased organ or tissue (e.g., non-diseased kidney or
liver) of a subject that has a disease (e.g., cancer or a tumor) in
a different tissue or organ can be assessed and compared to
vascular changes in a healthy subject. It should be appreciated
that one or more quantifications described herein (e.g., one or
more ratios of vascular changes in treated versus control organs or
tissue) can be used, either directly or indirectly, as a basis for
providing a quantitative assessment of vascular toxicity of a
particular compound or treatment.
[0255] Aspects of the technology described herein may be used to
study, identify, and or analyze geometrical, structural, and/or
distributional features of blood vessels that are associated with
one or more diseases or conditions represented by an animal of
interest. In some embodiments, an animal may be a disease model as
described herein. In some embodiments, an animal may be undergoing
a therapeutic regimen of interest. In some embodiments, an animal
may be treated with a candidate therapeutic compound. Accordingly,
aspects of the technology described herein may be used to identify,
analyze, and/or evaluate one or more vascular patterns or changes
in vascular patterns associated with a disease. Aspects of the
technology described herein also may be used to evaluate the
effects of one or more therapeutic regimens or candidate compounds.
In some embodiments, therapeutic effectiveness may be evaluated
using one or more vascular patterns or changes therein as a marker
of a response (or lack thereof) to treatment. Accordingly, aspects
of the technology described herein may be used to identify
particular vascular patterns that are indicative of certain
diseases or disease stages. These patterns can subsequently be used
in sensitive assays to detect diseases in vivo (e.g., in human
subjects). Other aspects of the technology described herein may be
used to select therapeutic regimens or candidate compounds for
administration to a patient (e.g., a human patient) in a
therapeutically effective amount and in a physiologically
acceptable form.
[0256] It should be appreciated that in some embodiments, an animal
(e.g., an animal that is perfused with a casting agent composition)
may be sacrificed prior to analysis regardless of whether the
analysis is performed in situ or not. Accordingly, in some
embodiments, changes over time may be studied using a plurality of
animals and using one or more animals for each time point of
interest. In some embodiments, different dosages, different
therapeutic regimens, different drugs or drug combinations, or any
combination of two or more thereof may be studied using different
animals (with at least one animal for each condition of interest).
It should be appreciated that combinations of time courses and
drugs, drugs dosages, or other therapeutic regimens similarly may
be studied using a plurality of different animals, each
representing a unique condition. It should be appreciated that the
different animals are preferably genetically identical or similar
(e.g., identical for at least one trait that is associated with a
disease or condition of interest). In some embodiments, the animals
may be mice, rats, sheep, cats, dogs, primates, or any suitable
non-human experimental animal.
[0257] In some embodiments, a combination of different drugs,
different doses, etc., may be evaluated at a series of time points
according to aspects of the technology described herein. Again, it
should be appreciated that a different animal may represent a
different drug, dosage, time point, or combination thereof, because
each animal may be sacrificed for analysis. However, in some
embodiments, a single animal may be tested at different sites
(representing, e.g., different drugs, dosages, time points, etc.)
depending on the impact of the casting agent that is used and the
site of administration of the casting agent.
[0258] In some embodiments, samples from one or more animals may be
prepared and analyzed periodically during the time course of a
treatment (e.g., using a group of animals exposed to the same
experimental conditions). In some embodiments, different conditions
may be compared. For example, separate groups of animals (e.g.,
groups of mice) may be exposed to a candidate drug and a placebo
(or other control). In some embodiments, subsets of animals (e.g.,
one or more animals) may be perfused with a casting agent
composition at different time points and vascular structures may be
imaged (e.g., directly or through reconstruction) for each time
point. For example, tumors may be induced in genetically-altered
mice using appropriate controls and different dose levels or
regimens (e.g., 1, 2, 3, 4, 5, or more different dose levels or
regimens) of one or more therapeutic compounds or compositions.
Vascular structures then may be analyzed at different time points
using methods of the technology described herein to evaluate the
effectiveness of a drug composition and/or to identify biological
markers that can be used to monitor a patient response to the drug
composition. It should be appreciated that vascular structures of
different sizes may be studied to identify structural features
and/or distribution patterns of interest. In some embodiments,
blood vessels having a diameter of about 50 microns are studied.
However, it should be appreciated that smaller or larger vessels,
or a combination thereof, may be studied.
[0259] In some embodiments, a vasculature characteristic may be
evaluated over time by comparing results at different time points.
However, it should be appreciated that the end-point of a study may
be used as a single time point and characteristics associated with
different diseases or treatments may be compared to identify or
infer changes associated with a disease, treatment, or other
condition of interest. Aspects of the technology described herein
can be used to analyze data obtained from any suitable image source
to identify one or more patterns associated with tubular structures
of different sizes (e.g., structural patterns of blood
micro-vessels). One or more parameters of a structural pattern can
be used as biomarkers for different biological conditions and
processes (including pathogenic conditions). Accordingly, aspects
of the technology described herein relate to disease detection,
diagnosis, grading, staging, disease monitoring, monitoring the
effectiveness of therapy and interventional applications based on
an analysis of structures (e.g., in situ structures) to identify
patterns that may be associated or correlated with a disease or
other physiological condition. According to aspects of the
technology described herein, a pattern may comprise one or more
different parameters. Parameters may be one or more structural
features of individual tubular structures and/or one or more
distribution properties (e.g., spatial distribution, spatial
orientation, frequency, number, etc., or any combination thereof)
of one or more tubular structures and/or one or more distribution
properties (e.g., spatial distribution, spatial orientation,
frequency, number, etc., or any combination thereof) of one or more
individual tubular structural features within a subject or a within
a region of interest in the subject, or any combination thereof.
Accordingly, a vasculature pattern may include one or more
structural features of an individual blood vessel (e.g.,
micro-vessels), a distribution of one or more blood vessels (e.g.,
micro-vessels) within a subject, a distribution of one or more
individual blood vessel structural features (e.g., individual
micro-vessel structural features), or any combination thereof. An
individual blood vessel structural feature may include, but is not
limited to, vessel tortuosity, curvature, branching (e.g.,
frequency, angle, hierarchy, etc.), diameter, direction, etc., or
any change (e.g., variation or frequency) of any of these features
over a predetermined length of the blood vessel being analyzed, or
any combination thereof. A distribution of blood vessels or
individual blood vessel structural features may include, but is not
limited to, a blood vessel density, a distribution of blood vessel
directions, a distribution of blood vessel diameters, a
distribution of distances between blood vessels, a distribution of
blood vessel spatial orientations (e.g., relative to each other), a
distribution of blood vessel curvatures, a distribution of any
other individual blood vessel structural features described herein,
other distributions of blood vessel parameters or any combination
of two or more thereof. It should be appreciated that the
distribution of blood vessels or blood vessel structural features
may be determined and/or analyzed for a predetermined region within
a subject (e.g., a target volume of tissue within a subject) or
within predetermined tissues or organs within a subject or
throughout the subject (e.g., within a vascular cast). It also
should be appreciated that either the absence or presence of blood
vessels or of individual blood vessel structural features within a
predetermined volume being analyzed may be a pattern parameter that
can be used in analytical methods of the technology described
herein. It also should be appreciated that one or more pattern
parameters may be monitored and/or analyzed as a function of time.
Accordingly, blood vessel patterns can be used as biomarkers for
different biological conditions and processes (including pathogenic
conditions). Accordingly, aspects of the technology described
herein relate to identifying and evaluating biological markers that
may be used for in vivo disease detection, diagnosis, grading,
staging, for disease monitoring, for monitoring the effectiveness
of therapy and interventional applications in live animals,
including humans, based on an analysis of vasculature patterns
including vasculature morphology and/or architecture in
experimental subjects, for example experimental animals (e.g.,
animals perfused with one or more casting agent compositions). In
one embodiment, the in vivo density, and/or diameter distribution,
and/or geometric orientation of blood vessels (e.g., micro-vessels)
may be analyzed, quantified, and/or evaluated for disease
detection, monitoring, and/or interventional applications. In one
embodiment, the sensitivity and specificity of disease diagnosis
may be enhanced by analyzing and evaluating in vivo vasculature
morphology and/or architecture associated with a tissue lesion.
Accordingly, aspects of the technology described herein include
detecting in vivo indicia of diseases associated with abnormal
vascular structures or patterns. Other aspects include disease
diagnosis, staging, grading, monitoring and prognosis, patient
treatment, drug development and validation, and research
applications. It should be appreciated that one or more biological
markers identified in vascular casts in association with a response
to a known drug or treatment may be used as a reference markers to
evaluate the effectiveness of additional drugs or treatments in
comparison to the known drug or treatment.
[0260] Certain embodiments according to aspects of the technology
described herein includes a method of analyzing geometric features
of blood vessels and correlating one or more features with a
biological process, condition, or disease. Accordingly, certain
geometric features of blood vessels may be used as biomarkers
indicative of particular biological processes, conditions, and/or
diseases.
[0261] In some embodiments, data for tubular structures (e.g.,
blood vessels) may been sorted into bins based on their size (e.g.,
their diameter). Aspects of the invention may increase the
analytical resolution when evaluating structural information that
is obtained for one or more experimental models and/or subjects
being evaluated. According to aspects of the technology described
herein, a binned structural analysis refers to any analysis of
tubular structures that have been sorted or categorized according
to size (e.g., according to the diameter or radius of the tubular
structure in an area of interest). For example, in some embodiments
a binned micro-vessel density (BMVD) analysis refers to an analysis
of blood vessel density based on blood vessels that have been
categorized according to vessel diameter in an area of
interest.
[0262] Binned analytical techniques can be applied to the analysis
of many different parameters that may be characteristic of tubular
structures. Binned analytical techniques may be performed on
tubular structures observed in casts or in vivo (e.g., in situ).
For example, bins of tubular structures having different diameters
can be evaluated to determine one or more of the following
parameters: tortuosity, curvature, density, branching frequency,
branching hierarchy (e.g., presence or absence of a branching
hierarchy), relative distribution and/or direction of tubular
structures (e.g., blood vessels), etc., or any combination thereof.
By performing the analysis on binned data, small changes that
primarily affect structures in one size range are more likely to be
detected, because they are not masked by a relative absence of
change in structures in other size ranges. Accordingly, methods of
the technology described herein can be used to refine an analysis
of tubular structures (e.g., blood vessels) over time or in
response to disease or treatment, etc., where the analysis may be
performed on casts and/or in vivo. Aspects of the technology
described herein can also be used to detect or delineate diseased
tissue (e.g., cancerous or pre-cancerous tissue, necrotic regions,
etc.) in casts and/or in vivo.
[0263] It should be appreciated that, regardless of the source of
information relating to vessel geometry, structure, and/or
distribution (e.g., from analysis of BMVD, casts, in vivo, images,
representations, etc., or any combination thereof), analytical
methods described herein may be used. Accordingly, any analytical
descriptions of vessel distributions that are provided in the
context of one source of information may be applied to that
analysis of vessel distributions obtained from one or more other
sources as appropriate.
[0264] In some embodiments, spatiotemporal information about the
vessel distribution provides numerous indicators about the health
of a tumor, the effectiveness of a treatment such as the efficacy
of a particular anti-angiogenic drug, and how a tumor is changing
over time with respect to differently sized vessels. Numerous
exemplary applications using one or more distribution analyses
(e.g., based on BMVD measurements), in accordance with various
aspects of the technology described herein. The inventors have
identified and disclosed various applications that are facilitated
by the acquisition of information about vessel characteristics,
distribution, size, shape, etc., in PCT application US2005/047081
filed on Dec. 22, 2005, which is hereby incorporated by reference
in its entirety. The inventors have appreciated that certain of
these applications are facilitated by obtaining one or more BMVD
measurements or by using one or more alternative binned analyses.
It should be appreciated that any application may involve an
analysis limited to one or more (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10,
or more) bins of microvasculature of different sizes. For example,
binned analyses may be useful for diagnostic applications. In one
embodiment, aspects of the technology described herein can be used
to detect and diagnose diseases associated with patterns (e.g.,
individual structural features or distributions) of in situ tubular
networks. In some cases, a diagnosis can be rendered from an
examination of the patterns (e.g., individual structural features
or distributions) of interest at a single time. Alternatively,
disease progression in a subject can be tracked by performing a
structural analysis at two or more (e.g., 3, 4, 5, 6, 7, 8, 9, 10,
or more) time points. Disease tracking can be used to provide
diagnostic and prognostic information for a patient. For example,
disease progression information can be used to assess the
aggressiveness and/or invasiveness of a tumor.
[0265] The technology described herein can be used to screen an
individual or a population for the presence of indicia relating to
one or more diseases. As mentioned herein, the screen may be a
whole body screen, or may be focused on one or more target regions
(e.g., specific organs or tissues).
[0266] In one embodiment, the techniques described herein can be
used automatically to identify individuals with one or more
disease-associated structural patterns or features. These
individuals can be subsequently tested for additional indicia of
disease. The subsequent testing can take any suitable form, as the
aspects of the technology described herein are not limited in this
respect. For example, follow on testing can employ conventional
techniques. As a non-limiting example, the use of aspects of the
technology described herein may enable cost-effective screening
techniques that may identify a relatively small pool of candidates
as at risk of a disease, and may justify the use of relatively more
expensive testing procedures to reach a final diagnosis or
prognosis, wherein the follow on techniques may be too expensive to
administer to a wider sample that has not been narrowed using the
techniques of the technology described herein. As a further
example, aspects of the technology described herein, either alone
or in combination with other techniques, can be used to perform
subsequent tests. In this respect, the sensitivity of the initial
screening can be set relatively high, such that it may indicate
some false positives, and subsequent application of techniques in
accordance with aspects of the technology described herein can be
employed with a higher degree of sensitivity that may provide more
detailed information.
[0267] In one embodiment, aspects of the technology described
herein can be used to screen a population of at risk individuals
(e.g., individuals with genetic or other risk factors for a disease
such as cancer, a circulatory disorder, or other disease) to
identify the presence of disease indicia in one or more
individuals.
[0268] In one embodiment, diagnostic methods of the technology
described herein are computer-implemented to increase efficiency
and throughput, and reduce variability associated with individual
physicians. However, as discussed herein, in some embodiments, the
final diagnosis may be made by a physician based on information
generated by an automated analysis or a structural representation
using aspects of the technology described herein.
[0269] As shall be appreciated from the foregoing, aspects of the
technology described herein can be used on patients known to have a
disease, or can be used to screen healthy subjects on a regular
basis. A subject can be screened for one or more diseases.
Screening can be done on a regular basis (e.g., weekly, monthly,
annually, or other time interval); or as a one-time event.
Different conditions can be screened for at different time
intervals and in function of different risk factors (e.g., age,
weight, gender, history of smoking, family history, genetic risks,
exposure to toxins and/or carcinogens etc., or a combination
thereof).
[0270] In one embodiment, aspects of the technology described
herein can be employed to diagnose, evaluate or stage diseases
associated with changes in vasculature structure. The detection of
small changes in vasculature structure may be informative for early
stage disease detection and disease monitoring. A morphological
determination of binned blood vessels may be analyzed and one or
more patterns (e.g., individual structural features or
distributions) may be evaluated for the presence of abnormal
properties. In one embodiment, a vasculature structure may be
obtained including a series of interconnected branched blood
vessels and may include arteries, arterioles, veins, venules,
capillaries, and other sized blood vessels. However, according to
aspects of the technology described herein, an interconnected
vasculature structure is not required and different sizes of blood
vessels can be analyzed separately and represented on a histogram
or other form of distribution representation. In some aspects of
the technology described herein, blood vessels of the entire body
can be analyzed, and in other aspects the blood vessels of a target
organ, tissue, or part thereof can be analyzed. In some aspects of
the technology described herein, only a subset of blood vessel
sizes is binned and analyzed (e.g., blood vessels with a diameter
below about 500 microns, preferably below about 200 microns, more
preferably below 100 microns, even more preferably below 50
microns, and even more preferably below 25 microns). In one
embodiment, only capillary blood vessels are analyzed. In another
embodiment, capillaries and small arteries and veins (e.g.,
arterioles and venules) are analyzed. For example, an arborescent
vasculature can be analyzed in any tissue where it is found (e.g.,
an arborescent mucosal vasculature such as the oesophageal
arborescent mucosal vasculature).
[0271] The branches of a vascular tree may be analyzed to glean
information about the status of the patient. In one embodiment, the
branches of a vascular tree may be followed to identify specific
regions where certain characteristics of angiogenesis may be
evaluated (e.g., start with a large branch and follow the tree to
second, third, or fourth, or subsequent levels of branching to
identify small blood vessels that may have abnormal structures if
they are providing a blood supply associated with a disease).
Alternatively, several different blood vessel sizes in the vascular
tree may be evaluated for signs of angiogenesis. In another
embodiment, the overall branching pattern of a vascular tree can be
analyzed. For example, a healthy vascular tree may be approximately
hierarchical in that the size of the blood vessels generally
decreases as the vessels branch. In contrast, a diseased (e.g.,
angiogenic) vascular tree may be less hierarchical with areas of
significant blood vessel branching with little or no decrease in
blood vessel size. It should be appreciated that the nature and
extent of the analysis may depend on the goal of the diagnostic
evaluation. For example, a full body scan can be evaluated
selecting all vascular structures and analyzing the entire vascular
network for signs of different diseases. Alternatively, a region of
a body suspected of being diseased may be selected and the data may
be processed to focus on the vasculature in that region (e.g., to
obtain a segmented representation of structures in the region of
interest). A region of interest may be an organ (e.g., pancreas,
liver, kidneys, breast, colon, etc.) or a tissue (e.g., skin
epidermal tissue, retinal tissue). The presence of an abnormal
vasculature structure can be an early indication of a range of
diseases for which early detection is critical for effective
treatment (e.g., retinal vascular changes are a common precedent to
the development of diabetes and hypertension).
[0272] Diseases associated with changes in vascular structure
(e.g., that can be detected by the presence of abnormal vascular
patterns at a given time or abnormal structural changes observed as
a function of time) include, but are not limited to, cancer, heart
diseases and related circulatory disorders, eye diseases, skin
disorders, and surgical conditions. For example, diseases and
conditions associated with changes in vascular structure include,
but are not limited to, tumor angiogenesis, recurrent and
progressive cancers, coronary artery disease, cardiomyopathy,
myocardial ischemia, arteriosclerosis, atherosclerosis,
atherosclerotic plaque neovascularization, arterial occlusive
disease, ischemia, ischemic or post-myocardial ischemia
revascularization, peripheral vascular disease (including diabetic
retinopathy), thromboembolic diseases (e.g., stroke, pulmonary
embolism, brain aneurisms, and deep venous thrombosis),
claudication, rheumatologic disorders (e.g., arthritis), immune
disorders (e.g., rheumatoid arthritis, vasculitis, Wegner's
granulomatosis, and systemic lupus erythematosis (SLE)), pulmonary
disorders (including, emphysema, COPD, idiopathic pulmonary
fibrosis, pulmonary arterial hypertension, and other respiratory
disorders), myeloma, vascular proliferative disorders,
gastrointestinal disorders (e.g., Crohn's disease, ulcerative
colitis, and inflammatory bowel disease (IBD)), gynecologic
disorders (endometrial polyp, vaginal bleeding, endometriosis,
dysfunctional uterine bleeding, ovarian hyperstimulation syndrome,
preeclempsia, polycystic ovarian syndrome (PCO), cervical cancer,
and cervical dysplasia), skin disorders (infantile hemangioma,
verruca vulgaris, psoriasis, neurofibromatosis, epidermolysis
bullosa, Stevens-Johnson syndrome, and toxic epidermal necrolysis
(TEN)), eye disorders (macular degeneration, maculopathies,
diabetic retinopathy, and retinopathy of prematurity (retrolental
fibroplasia)) wound healing, inflammation associated with immune
responses, ischemia including limb ischemia and cardiac ischemia,
Alzheimer's disease and other disorders such as wound dehiscence,
Buerger Disease (thromboangitis obliterans, arteriosclerosis
obliterans (ASO), ischemic ulcers) multiple sclerosis, idiopathic
pulmonary fibrosis, HIV infections, plantar fasciosis, plantar
fasciitis, Von Hippel-Lindau Disease, CNS hemangioblastoma, retinal
hemangioblastoma, thyroiditis, benign prostatic hypertrophy,
glomerulonephritis, ectopic bone formation, and keloids.
[0273] These different diseases are characterized by different
changes in vasculature structure. Accordingly, in one aspect of the
technology described herein, parameters and scoring methodologies
are used to detect, diagnose, and monitor particular diseases and
their related therapies based upon particular characteristics of
vasculature structure indicative of the disease. Even within each
disease category, different diseases can be characterized by
different changes in vasculature structure. Accordingly, structure
mining and scoring can be fine-tuned to increase the sensitivity
for particular types of disease within a category (e.g., lung
cancer score, breast cancer score, etc., can be developed).
Patient-specific scoring parameters can also be developed to follow
the progression of a specific disease or disorder in a patient.
[0274] Structural vasculature changes include changes in vascular
architecture and vascular morphology affecting blood vessels and/or
lymph vessels. Structural changes can involve neovascularization
(including the growth of large blood vessels (e.g., arteriogenesis)
and the growth of microvasculature (angiogenesis)), large blood
vessel expansion, and vascular necrosis. Angiogenesis involves the
formation of new blood vessels that sprout from preexisting blood
vessels. Angiogenesis is different from vasculogenesis, which is
the de novo formation of vessels that occurs primarily during
development. Vasculogenesis is rarely associated with a disease or
disorder. However, aspects of the technology described herein can
be used to study the natural process of vasculogenesis to help
identify and understand defects in de novo blood vessel
formation.
[0275] Angiogenesis is often associated with tumor growth and is a
useful biomarker for cancer. Angiogenesis also can be associated
with conditions where new blood vessel growth occurs in response to
a reduced oxygen supply or blood flow (whether due to thrombosis,
embolism, atherosclerosis, or other chronic occlusion or narrowing
of the vasculature). Certain respiratory, cardiovascular, and
inflammatory disorders also are associated with angiogenesis.
[0276] Angiogenic blood vessels have structural characteristics
that are different from those of established blood vessels. For
example, the branching patterns and tortuosity of angiogenic blood
vessels are very different from those of normal blood vessels.
These and other structural features are found predominantly in
microvasculature and can be used for mining and scoring vasculature
structural images. However, changes in larger blood vessels such as
arteries and veins also may be associated with certain diseases or
disease stages (e.g., growth and development of large tumors or
late-stage tumors).
[0277] The vasculature that supports a tumor is typically
associated with the connective tissue of the tumor (the stroma)
that supports the malignant cells (in the parenchyma). As discussed
herein, tumor blood vessels are irregularly spaced and
characterized by heterogeneous structural patterns or features.
However, the formation of tumor blood vessels and other forms of
angiogenesis may involve a series of characteristic stages (see,
for example, Dvorak, 2003, American Journal of Pathology, Vol.
162:6, pp. 1747-1757, the disclosure of which is incorporated
herein by reference in its entirety). Early stage angiogenesis may
be characterized by vascular hyper-permeability, fibrin deposition
and gel formation, and edema. This may result in the enlargement of
micro-vessels such as venules. The cross-sectional area of an
enlarged micro-vessel may be about 4 fold that of a normal
micro-vessel. The perimeter of an enlarged micro-vessel may be
about 2 fold that of a normal micro-vessel. Enlarged micro-vessels
may occupy about 4-7 fold the volume of normal micro-vessels in a
region of active angiogenesis. The appearance of enlarged
micro-vessels may be followed by the appearance of "mother" vessels
that are enlarged, thin-walled, serpentine, and hyper-permeable.
Mother vessels may undergo a process of bridging whereby
trans-luminal bridges are formed dividing the blood flow within the
vessel into smaller channels. A developing mother vessel also may
contain one or more glomerular bodies that may expand to divide the
lumen of the mother vessel into several smaller channels that are
typically tortuous. Bridging and glomerular body formation in
mother vessels may lead to the appearance of small capillaries
characteristic of angiogenesis. However, certain mother vessels
persist as abnormally enlarged vessels with thin walls. These
vascular malformations are often characterized by the presence of
an asymmetric muscular coat and perivascular fibrosis. Small
arteries and arterioles also may increase in size in diseased
tissue. Aspects of the technology described herein include
detecting and/or monitoring any one or more of the blood vessel
structural changes described herein. In one embodiment, the
presence of one or more patterns (e.g., individual structural
features or distributions) characteristic of new blood vessel
formation may be used to detect or monitor a disease. In another
embodiment, the presence of one or more specific patterns (e.g.,
individual structural features or distributions) may be used to
determine the stage of angiogenesis (e.g., early-stage, mid-stage,
late-stage, etc.) in a body region.
[0278] Accordingly, abnormal changes in blood vessel size (diameter
and/or length) can be early signs of diseases such as cancer or
other disease associated with an increased blood supply. Changes in
blood vessel size may occur before any structural signs of
angiogenesis appear. In one embodiment, aspects of the technology
described herein are useful to detect blood vessels (e.g.,
capillaries) that are swollen and/or longer than normal. For
example, aspects of the technology described herein are useful to
detect abnormally long intrapapillary capillary loops in situ
(e.g., associated with early stages of cancer in oesophageal
mucosa).
[0279] In some embodiments, blood vessel changes indicative of
necrosis in tumor tissues may be indicative of the aggressiveness
of the tumor tissue and/or the likelihood of metastasis, and/or the
responsiveness to therapy, and/or the efficacy of a therapeutic
treatment (e.g., a candidate drug), and/or an therapeutic treatment
selection and/or modification (e.g., a change in drug or dose for
an individual patient). Accordingly, in situ patterns (e.g.,
individual structural features or distributions) indicative of
necrosis may be useful biomarkers for patient prognosis. In certain
embodiments, necrosis within a region of a tumor may be indicated
by one or more of the following patterns (e.g., individual
structural features or distributions) within that region: a
collapse in blood vessel structure, poor vascularization (e.g., a
low blood vessel density relative to other regions of the tumor or
relative to the perimeter of the tumor), a change in blood vessel
size or shape over time, a lower than threshold number of blood
vessels, blood vessels (e.g., in the microvasculature or the
capillaries) that are separated by a greater than threshold
distance (e.g., by more than 100 microns, more than 150 microns, or
more than 200 microns) within a volume of the tumor, micro-vessel
diameter and/or density indicative of undervascularization, etc.,
or any combination thereof. In some embodiments, a volume of
avascularization or undervascularization may be evaluated or
quantified and used as an indicator of necrosis. It should be
appreciated that other indicia of necrosis may be used, alone or in
combination with blood vessel features. Other indicia may include
indicia of tissue collapse or cavitation that may be visualized
(e.g., using CT etc.) and/or indicia of tissue viability using one
or more markers of metabolic activity (e.g., ones that may be
analyzed using a PET scan, etc.). One or more reference indicia
(e.g., a reference volume of avascularization or
undervascularization may be identified by analyzing vascular casts
of necrotic tumor tissue (e.g., in a xenograft tumor model, for
example in an orthotopic or an ectopic tumor xenograft).
[0280] Aspects of the technology described herein may be used for
the detection (e.g., the automatic detection) of necrotic areas in
a subject (e.g., in a tumor in a subject). A necrotic region is an
avascular region within the boundary of a diseased tissue. Methods
of the technology described herein may be used to detect (e.g.,
automatically) the transition between the vascularized diseased
tissue and avascular region that defines the boundary of the
necrotic region.
[0281] Aspects of the technology described herein also may be used
to detect or evaluate (e.g., automatically) a response to therapy.
For example, a response to therapy (e.g., to a specific drug and/or
a specific dosage of a drug, and/or to a combination of drugs and
specific dosages of these drugs, etc.) can be detected and assessed
as follows. Changes in the vascular patterns (e.g. vessel
normalization/straightening, disappearance of smaller diameter
vessels leading to lower micro-vessel density and to skewing of the
vessel diameter distribution towards the larger vessels) may be
detected and/or evaluated within the volume defined by the boundary
of the diseased tissue and the boundary of the necrotic area. An
increase in the absolute volume size of the necrotic area and/or
the rate of such change while the total volume of the disease (e.g.
tumor) volume stays constant may be detected and/or evaluated as an
indicator that the therapy is effective. An increase in the ratio
between the absolute volume size of the necrotic area and the total
disease (e.g., tumor) volume and/or the rate of change in this
ratio may be detected and/or evaluated and used as an indicator
that the therapy is effective. A ratio of the diseased tissue
volume and the necrotic region volume may be detected and/or
evaluated and when it approaches 1 and the overall diseased tissue
volume starts shrinking it provides an indication that a therapy is
effective. In some embodiments, reference indicia may be obtained
from analyzing casts (e.g., appropriate vascular casts). However,
reference indicia may be obtained from any suitable data relating
to blood vessel structures (e.g., view data, scan data, in vivo
data, etc., or any combination thereof).
[0282] Structural representations of blood vessels can be mined to
identify and evaluate certain patterns (e.g., individual structural
features or distributions) that can be used to provide a score that
is related to the probability that the blood vessels are normal or
abnormal (e.g., disease associated). Accordingly, in some
embodiments a binned analysis may be predictive of a response to
therapy.
[0283] In certain embodiments, a binned analysis may be sensitive
to vasculature changes resulting from unwanted side-effects
associated with one or more therapeutic drugs. Accordingly, binned
analysis may be used to detect or quantify toxic side-effects of
certain drugs.
[0284] The morphology of blood vessels (e.g., binned blood vessels)
can be mined to identify and evaluate certain patterns (e.g.,
individual structural features or distributions) that can be used
to provide a score that is related to the probability that the
blood vessels are normal or abnormal (e.g., disease associated).
Patterns (e.g., individual structural features or distributions)
for scoring blood vessels include, but are not limited to, the
following: diameter, curvature, tortuosity (including, for example,
the degree of tortuosity, the length of the blood vessel along
which abnormal tortuosity is observed, etc.), variability or
heterogeneity (including spatial variability or heterogeneity over
distance or in a volume), branching shape or pattern, branching
density, branching hierarchy, blood vessel density, distribution of
vessel size (ratio of microvasculature to macrovasculature) a field
effect (the presence of blood vessels bending towards a specific
region), blood vessel diameter distribution, variability of the
geometric orientation of blood vessels or fragments thereof, and
the distribution of the orientation(s) within a field. The score
may have more significance if two or more (e.g., 3, 4, 5, 6, 7, 8,
9, 10, or more, or all) of these parameters are evaluated. In some
embodiments, a score is generated using one or more of these
structural parameters combined with additional information such as
patient-specific medical information (e.g., age, weight, height,
gender, etc.) and the presence of one or more additional indicators
of disease such as a visible lesion on an X-ray or other image. In
some embodiments, a score can be provided for a tumor. An example
of a useful score is one that reflects the vascularity of a tumor.
An abnormally high vascularity (measured as a higher than normal
blood vessel number, density, length, or combination of the above)
is generally indicative of a more aggressive or invasive tumor. In
one embodiment, vascularity is evaluated by measuring the volume of
the lumen of angiogenic vasculature (the volume within the blood
vessel tree associated with a tumor). In another embodiment, a
measure of vascularity is provided by dividing the volume of the
angiogenic lumen by the volume of the solid tumor. Additional
information can be gleaned from obtaining a score (or other
structural evaluation) at two or more times. A changing score (or
other structural evaluation) is indicative of an evolving
vasculature that could be associated with a disease or disorder. It
should be appreciated that the patterns (e.g., individual
structural features or distributions) described herein can be
identified and analyzed for a field of analysis without imposing a
connectivity on the vessels being studied. In some embodiments, it
may be sufficient to analyze only fragments of blood vessels in
order to detect one or more structural features of individual
vessels or geometrical features of a field of vessels that are
different from normal features. For example, blood vessel fragments
having an average length of 0.5 mm, 1 mm, 5 mm, 10 mm, 50 mm, 1 cm,
5 cm, 10 cm, 50 cm, etc. may be used. However, it should be
appreciated that shorter or longer or intermediate lengths may be
used. The scoring and mining aspects of the technology described
herein can be automated. Accordingly, diseased (e.g., angiogenic)
vasculature can be automatically detected amidst normal
vasculature. Various vasculature parameters can be automatically
detected and scored, either separately or in any combination,
including vessel tortuosity, vessel branching, vessel density, and
total intra-vascular volume, but the technology described herein is
not limited to any particular parameter or combination.
[0285] In one embodiment, aspects of the technology described
herein can be used to detect blocked blood vessels, and
thromboembolic events, including stroke, lung emboli, blocked
micro-coronaries, deep-vein thrombosis, etc. Blocked blood vessels
can be detected (1) directly by detecting structural changes in the
blocked blood vessel (e.g., detecting a clot, wall thickening, or
other signs of reduced flow) and/or (2) indirectly by detecting new
vasculature that was generated in response to the blockage. In
general, the formation of collateral blood vessels is more ordered
than angiogenesis associated with cancer. One aspect of the
technology described herein described herein also allows clots to
be detected in small blood vessels.
[0286] As discussed herein, aspects of the technology described
herein can be used to screen the entire vasculature structure of a
human or other animal to screen for any form of abnormality in any
tissue. Alternatively, a subset of the body may be screened.
Accordingly, the structures of binned vessels can be analyzed for
one or more organs or tissue types. In addition, only a portion of
the vessels in any predetermined bin may be analyzed within any
target volume as opposed to the entire vascular tree in that
volume. This may be done by analyzing structure data focused on the
area of interest, or large amounts of structure data may be
obtained, but an analysis may be restricted to a subset of the
available data. In some embodiments, only a portion of a vascular
tree may be binned and/or analyzed, for example only a portion of
those vessels that are of a particular size range. In some
embodiments, only fragments of a vascular tree are represented
and/or analyzed if the fragments are sufficiently informative to
provide patterns (e.g., individual structural features or
distributions) of interest. Fragments may include branches or may
be unbranched. The portion of the vasculature being analyzed may be
statistically significant, such that any observation (normal or
abnormal) is physiologically significant. For example, branched
structures may not be required for the analysis if a sufficient
number of vessel substructures are analyzed to confidently detect
any other patterns (e.g., individual structural features or
distributions) that may be associated with vasculature changes
(e.g., angiogenesis) such as high vessel density. In aspects of the
technology described herein, vascular patterns may be detected
and/or evaluated in situ in a volume of 1 mm.sup.3, 2 mm.sup.3, 5
mm.sup.3, 1 cm.sup.3, 2 cm.sup.3, 5 cm.sup.3, 10 cm.sup.3, etc.
However, smaller or larger or intermediate volumes also may be
analyzed. In some embodiments, vascular patterns or structures are
evaluated over an entire model tissue or organ (e.g., for an entire
orthotopic or ectopic tumor model).
[0287] Different tissues and organs have different and
characteristic blood vessel patterns (e.g., the lung which is
highly vascularized). Accordingly, in one embodiment, structural
analyses and associated structural parameters may be optimized for
evaluating different tissues.
[0288] In some embodiments, scan data is obtained and/or analyzed
for one or more organs (e.g., lung, heart, colon, brain, liver,
pancreas, kidney, breast, prostate, etc.) or tissue (e.g., skin,
bone, etc.) or portion of any of the above.
[0289] Brains may be evaluated for signs of brain tumors and/or
other neurological disorders that can be associated with changes in
vascular patterns. For example, Alzheimer's may be associated with
certain vascular abnormalities. In one embodiment, one or more
changes in blood vessel pattern (e.g., shape and/or size) may be
detected as an indicator of high blood pressure in the brain.
[0290] In some embodiments, certain specific regions of organs or
tissues are focused on. For example, atherosclerosis is typically
found in certain parts of the arterial tree (e.g., bifurcations,
side branches, regions opposite flow dividers, and other areas
where angiogenesis often occurs in association with
atherosclerosis) and certain cancers tend to occur more frequently
in certain organ or tissue regions (e.g., colon cancers are not
distributed evenly along the length of the colon).
[0291] In other embodiments, aspects of the technology described
herein may be used to follow up with individuals who have been
identified as having one or more other indicia of disease (e.g.,
fecal occult blood, a colon polyp, a lung nodule, one or more cysts
or other indicia of disease). Aspects of the technology described
herein may be used to confirm the presence of a disease, determine
a location for the disease-associated lesion, or provide an
evaluation or prognosis of a disease. For example, aspects of the
technology described herein may be used to determine whether
abnormal vasculature is present at the site of a lesion (e.g. a
colon polyp, a lung nodule, a bladder cyst, a prostate cyst, a
breast cyst, a spot on a mammography, or any other cyst, lump, or
spot that may be detected physically, visually, or using any other
diagnostic technique) and help evaluate the likelihood of a
malignancy (or other carcinogenic disease stage) associated with
the lesion. Accordingly, aspects of the technology described herein
may be used for virtual malignancy detection (e.g., virtual
colonoscopy, virtual colon malignancy detection, virtual
bronchoscopy, virtual lung malignancy detection, virtual
mammography, virtual cystoscopy, etc.).
[0292] In other embodiments, aspects of the technology described
herein may be used for screening a cancer patient to evaluate the
extent of a cancerous lesion and/or to screen for the presence of
one or more metastatic lesions (e.g., one or more loci associated
with angiogenesis). A cancer patient may be screened upon initial
diagnosis of a primary cancer. In addition or alternatively, a
cancer patient may be screened at least once after an initial
cancer treatment (e.g., surgery, radiation, and/or chemotherapy).
This screening may include the original cancer locus to detect any
cancer recurrence. This screening may include similar body tissue
to screen for the presence of other lesions in the same tissue or
organ (e.g., the entire colon may be screened when a cancerous
lesion is detected in one region of the colon, the second breast
may be screened when a cancerous lesion is detected in one breast,
etc.). This screening also may be extended to the whole body or to
one or more other loci suspected of containing a metastatic lesion.
In one embodiment, a cancer patient may be screened several times
after an initial cancer treatment (e.g., at time intervals of about
6 months, about 1 year, about 2 years, about 5 years, or at other
time intervals).
[0293] In one embodiment, a follow up procedure may involve
screening one or more organs or tissues for the presence of a
metastatic lesion. Different cancers may have different
characteristic patterns of metastasis. Accordingly, different
target loci may be screened for different cancers. For example,
metastatic breast cancer typically spreads to the lungs, the liver,
bone, and/or the CNS. Therefore, one or more of these tissue types
or organs may be screened after a patient is diagnosed with breast
cancer. Similarly, other target loci may be screened after a
patient is diagnosed with another cancer type. In some embodiments,
the entire body of a cancer patient may be screened for indicia of
metastasis.
[0294] In one aspect, an initial screen may be performed on an
entire body, or an entire organ, using a low resolution
representation and/or, for example, analyzing only one or two or a
small number (e.g., less than five) pattern parameters in order to
detect indicia of a disease. Subsequently, the presence and or
nature of the disease may be diagnosed using a higher resolution
representation and/or, for example, analyzing one or more
additional pattern parameters or alternative pattern parameters
than those that were analyzed for the initial detection.
[0295] In some embodiments, small changes in blood vessel
distributions may be observed (for example as measured by a ratio
between the number of blood vessels of two or more different sizes
in a region of interest, for example, a tumor in an animal model)
and used as a biomarker. Such biomarkers may represent early
changes (e.g., early changes in tumor growth or response to
therapy) that occur before later changes in tumor size and/or tumor
morphology. It should be appreciated that some or all of the
diagnostic aspects of the technology described herein can be
automated as described herein.
[0296] It should be appreciated that some or all of the diagnostic
aspects of the technology described herein can be automated as
described herein.
[0297] Aspects of the technology described herein also can be used
to identify the location of a disease by locating one or more
structural abnormalities associated with the disease. This
information can be used to target a biopsy procedure or a treatment
(e.g., a treatment with one or more toxic chemicals, radiation,
heat, cold, small molecules, gene therapy, surgery, any other
treatment, or a combination of two or more of the above) to the
precise location of a disease lesion, or for any other purpose.
[0298] In one embodiment, an imaging device is connected to a
computer that provides a real-time visual display of the disease
lesion. In one embodiment, a real-time visual display may be an
accurate model of a body region and lesion along with associated
vasculature (as opposed to an actual image). This visual
information can be used to guide a surgical instrument for a
biopsy. Alternatively, the information can be used to guide an
invasive (e.g., surgical removal or bypass) or non-invasive (e.g.,
radiation) treatment procedure to the site of the disease lesion
(e.g., tumor or blood clot).
[0299] In some embodiments, aspects of the technology described
herein may be used to define the boundary between diseased and
non-diseased tissues, or between necrotic and non-necrotic tissue,
etc., or any combination thereof. For example, a boundary may be
identified or defined by analyzing binned data for several areas of
interest and identifying adjacent areas having very different blood
vessel densities (or differences in other morphological parameters
that are associated with disease, necrosis, etc., or any
combination thereof.
[0300] In one embodiment, aspects of the technology described
herein may be used to identify an area of tissue for treatment
before the treatment is applied. For example, a treatment target
region may be identified by detecting a boundary of chaotic blood
vessel structures. The area may be assessed after treatment to
confirm that the treatment was appropriately targeted. In one
embodiment, a structure may be analyzed pre-operatively to identify
the extent of tissue to be removed from a body region. In one
embodiment, a body region may be analyzed post-operatively to
determine whether any abnormal structures were missed. This may be
used to confirm the success of a radiation treatment or a surgical
removal of diseased tissue. Alternatively, this may be used to
decide on further surgery and/or another form of treatment. In
another embodiment, a disease boundary may be defined or depicted
by the boundary of abnormal vasculature. A treatment (e.g.,
radiation therapy, surgery, etc.) may be guided by and/or
restricted to a volume encompassed by the disease boundary.
[0301] In one embodiment, aspects of the technology described
herein can be used to evaluate the success of a surgical implant or
transplant. For example, aspects of the technology described herein
can be used to evaluate the formation of new blood vessels after an
organ or tissue transplant.
[0302] In another embodiment, the development of new blood vessels
may be monitored after removal of tumor tissue or after a tumor
biopsy, both of which may trigger angiogenesis and/or convert a
dormant tumor into a malignant tumor.
[0303] It should be appreciated that some or all of the
interventional aspects of the technology described herein can be
automated as described herein.
[0304] Aspects of the technology described herein also can be used
to optimize a therapeutic treatment for a patient. The extent of
disease progression or regression can be monitored in response to
different treatment types or dosages, and an optimal treatment can
be identified. The optimal treatment may change as the disease
progresses. The effectiveness of the treatment over time can be
monitored by analyzing changes in disease-associated patterns
(e.g., individual structural features or distributions) using the
aspects of the technology described herein described herein.
[0305] In one embodiment, a first therapy can be administered and
its effectiveness on slowing, stopping, or reversing abnormal blood
vessel growth can be monitored either irregularly or at certain
time intervals (e.g., daily, weekly, monthly, or other time
intervals). In some embodiments, if a first therapeutic regimen
does not have a desired effect on disease progression, a second
therapeutic regimen can be evaluated. Similarly, additional
therapeutic regimens can be evaluated on a patient-by-patient
basis. Additionally, the technology described herein can be used to
optimize a chosen therapeutic regimen (e.g., optimize dosage,
timing, delivery, or other characteristic of a drug or other
treatment) by monitoring the effect of minor therapeutic changes
and using the conditions that appear to be most effective for the
condition and the patient.
[0306] When looking at the therapeutic effectiveness of a
treatment, disease-specific parameters may be monitored. Of course,
all parameters can be obtained and only a subset reviewed. However,
it may be more efficient to simply obtain binned data only for
those parameters that characterize the disease.
[0307] According to aspects of the technology described herein,
patterns (e.g., individual structural features or distributions)
that are used to detect angiogenic vasculature and other abnormal
blood vessels also can be used to monitor a disease response to
treatment. For example, the total vascularity or any other
volumetric analysis of angiogenic or other diseased vasculature,
and the distribution of vessel size (e.g., a ratio of small to
large blood vessels) can be used independently or together as
indicators of disease progression or regression. In general,
microvasculature disappears before macrovasculature if an
anti-angiogenic treatment (or other disease treatment) is
effective. Therefore, an effective treatment results in a shift in
the distribution of blood vessel sizes towards larger vessels. An
index of anti-angiogenic activity can be scored as either a loss of
small blood vessels or a shift of observed blood vessels towards a
single size (or both).
[0308] In another aspect, the parameters can be (or include)
changes over time. For example, a structure present at a second
time can be compared to a structure present at a first time. In one
embodiment, a disease may be tracked pre-therapy and/or
post-therapy. Naturally, additional time points can be used. The
time points may depend on the condition being observed (e.g., is it
the progression of a disease that is already identified, is it the
screening of patient(s) over time). Time periods can be daily,
weekly, monthly, annual, or shorter, intermediate or longer time
periods. Time intervals may be a series of regular time periods.
However, other time intervals may also be useful. In one
embodiment, a patient-specific baseline is established and
monitored over time. For example, vasculature changes in the colon,
breast, or other tissue or organ can be monitored periodically.
[0309] In one aspect of the technology described herein, a type of
treatment may be determined by the degree or extent of abnormal
vascular structures (e.g., angiogenesis) that is detected at one or
more suspected disease loci (e.g., cancerous loci). For example, if
a suspected cancerous locus or metastasis is pre-angiogenic or
associated with early stage angiogenesis, it may be appropriate to
monitor the locus without any form of treatment. However, an
appropriate therapy may involve the administration of one or more
angiogenesis inhibitors to prevent the formation of any new
vasculature. If a suspected cancerous locus or metastasis is
associated with mid-stage angiogenesis, an appropriate therapy may
be the administration of one or more angiogenesis inhibitors. A
patient with mid-stage angiogenesis at a suspected locus also
should be monitored so that any further blood vessel development
can be treated more aggressively. If a suspected cancerous locus or
metastasis is associated with late stage angiogenesis, an
appropriate treatment may involve at least one or more of
chemotherapy (e.g., cytotoxic chemotherapy and/or hormone-based
chemotherapy), radiation, surgery, and/or treatment with one or
more angiogenesis inhibitors. However, it should be appreciated
that any of the above treatment options may be used to treat a
patient with any one or more lesions associated with any degree of
angiogenesis.
[0310] Examples of angiogenesis inhibitors include but are not
limited to 2-methoxyestradiol (2-ME), AG3340, Angiostatin,
Angiozyme, Antithrombin III, VEGF inhibitors (e.g., Anti-VEGF
antibody), Batimastat, bevacizumab (avastatin), BMS-275291, CAI,
2C3, HuMV833 Canstatin, Captopril, Cartilage Derived Inhibitor
(CDI), CC-5013, Celecoxib (CELEBREX.RTM.), COL-3, Combretastatin,
Combretastatin A4 Phosphate, Dalteparin (FRAGIN.RTM.), EMD 121974
(Cilengitide), Endostatin, Erlotinib (TARCEVA.RTM.), gefitinib
(Iressa), Genistein, Halofuginone Hydrobromide (TEMPOSTATIN.TM.),
Id1, Id3, IM862, imatinib mesylate, IMC-IC11 Inducible protein 10,
Interferon-alpha, Interleukin 12, Lavendustin A, LY317615 or AE-941
(NEOVASTAT.TM.), Marimastat, Maspin, Medroxpregesterone Acetate,
Meth-1, Meth-2, Neovastat, Osteopontin cleaved product, PEX,
Pigment epithelium growth factor (PEGF), Platelet factor 4,
Prolactin fragment, Proliferin-related protein (PRP), PTK787/ZK
222584, ZD6474, Recombinant human platelet factor 4 (rPF4), Restin,
Squalamine, SU5416, SU6668, SU11248 Suramin, Taxol, Tecogalan,
Thalidomide, Thrombospondin, TNP-470, TroponinI, Vasostatin, VEG1,
VEGF-Trap, and ZD6474.
[0311] Some embodiments may include a method of selecting a subject
for treatment and/or selecting a treatment or a course of therapy
based on the analysis of certain in situ vascular structures. A
method may involve analyzing in situ vascular structure(s) in a
human subject to obtain, for example, a score. The score may be
compared to a control score (e.g., in an apparently healthy
population) or to a previous score from a previous analysis on the
same subject. The treatment or the course of therapy may be based
on such a comparison. In some embodiments, obtaining an analysis of
vascular structures is repeated so as to monitor the human
subject's response to therapy over time. In some embodiments of
this aspect of the technology described herein, the method further
comprises measuring a second index of disease in the human subject
wherein deciding on the treatment or course of therapy is also
based upon the measurement of said second index.
[0312] In certain embodiments, patients having a tumor that is
under-vascularized (e.g., one that shows signs of necrosis) may be
selected for treatment with one or more anti-angiogenic compounds.
Under-vascularized tumors may be identified as those that have a
low density of blood vessels, or for which the blood vessel
diameters are low (e.g., below a threshold number typical of
vascularized tumors).
[0313] Aspects of the technology described herein also may include
monitoring the effectiveness of a therapy by monitoring the
presence of blood vessel patterns or features over time. For
example, the progressive loss of blood vessels in a tumor in
response to treatment may be a sign that a therapy is effective. In
contrast, the absence of any impact on vascularization may be an
indicator that a treatment is not being effective in a patient and
that an alternative therapy should be considered or used.
[0314] It should be appreciated that some or all of the therapeutic
aspects of the technology described herein can be automated as
described herein.
[0315] In one embodiment, aspects of the technology described
herein can be used to understand structural changes associated with
biological processes of interest (e.g., disease development and
progression). For example, an animal's vasculature can be analyzed
to identify additional patterns (e.g., individual structural
features or distributions or changes associated only with certain
binned size ranges) that may be associated with wound healing or
different diseases or different disease stages. These additional
patterns (e.g., individual structural features or distributions)
may be used in one of more of the diagnostic, intervention,
therapeutic, and development aspects of the technology described
herein.
[0316] In one embodiment, aspects of the technology described
herein can be used to understand structural changes associated with
medical procedures. For example, an animal's vasculature can be
analyzed to identify changes associated with post-surgical wound
healing or implant/transplant (including xenografts) growth or
rejection.
[0317] It should be appreciated that some or all of the research
aspects of the technology described herein can be automated as
described herein.
[0318] In another embodiment, aspects of the technology described
herein can be used in screens of compound libraries or to validate
candidate compounds for treating diseases associated with abnormal
internal structures (e.g., abnormal tubular networks). Aspects of
the technology described herein allow efficient high throughput
analyses of internal structural changes using binned data (e.g.,
BMVD). These changes can act as surrogate markers (biomarkers) for
certain diseases. As a result, the screening process can be
automated to a large extent, and the time for obtaining results
significantly shortened when compared to current validations that
often involve waiting for disease symptoms to change and also may
require tissue biopsies.
[0319] Aspects of the technology described herein may be used for
identifying and quantifying vascular patterns (e.g., structural
features) that can be used as surrogate markers for diagnostic,
therapeutic, and research and development purposes. Surrogate
markers are useful for reducing the time of diagnosis, therapy
evaluation, and drug development. A surrogate marker can be used as
an early indicator for disease diagnosis, disease prognosis, or
drug effectiveness, without waiting for a clinical outcome (e.g.,
increased survival time in response to a drug). So, a vasculature
analysis can be used as a surrogate marker for drug development (in
both pre-clinical and clinical trials), for clinical screening
(e.g., breast, lung, or colon screening), and for clinical therapy
monitoring. For example, binned vasculature structure may be a
useful surrogate marker for angiogenesis related diseases such as
cancer.
[0320] In one embodiment, aspects of the technology described
herein provide methods for screening and/or validating candidate
compounds or therapies for their effectiveness in treating
neo-vasculature formation and/or vasculature pattern changes
associated with disease. Aspects of the technology described herein
may be used to evaluate individual or small numbers of compounds or
to screen libraries to evaluate and/or identify a plurality of
candidate compounds (e.g., by administering these compounds,
individually or in groups, to an experimental animal such as a
mouse and evaluating their effect on angiogenic vasculature).
Libraries may contain any number of compounds (e.g., from
approximately 100 to approximately 1,000,000) Different types of
compounds can be screened, including antibodies, small molecules,
etc., or any combination thereof. However, the technology described
herein is not limited by the number and/or type of compounds that
can be evaluated.
[0321] In one embodiment, the effectiveness of a candidate compound
can be compared to a reference compound. A reference compound can
be any compound with a known effect on a structure. For example,
Avastin (Genentech) is a known monoclonal antibody against vascular
endothelial growth factor (VEGF) that can be used as a reference to
test the effect of a candidate compound on neovasculature growth.
Other examples of compounds include, but are not limited to, Sutent
and Nexavar.
[0322] It should be appreciated that some or all of the development
aspects of the technology described herein can be automated as
described herein.
[0323] It also should be appreciated that any one or more
geometrical, structural, and/or distributional parameters described
herein may be evaluated by comparison to a reference parameter. In
some embodiments, a reference parameter may be an amount or score
for that parameter in a normal or healthy subject. In other
embodiments, a reference may represent a diseased condition. In
some embodiments, a change or amount of any structural parameter
that is correlated or associated with a disease or condition as
described herein may be a statistically significant change or
difference in that parameter in a diseased or test subject relative
to a reference subject. In some embodiments, a difference or change
in a structural parameter may be an increase or a decrease in a
particular parameter (or a combination of parameters). An increase
in a parameter may be at least a 5%, 10%, 20%, 30%, 40%, 50%, 60%,
70%, 80%, 90%, 100%, or greater increase in that parameter in a
test subject relative to a reference subject. Similarly, a decrease
in that parameter may be at least a 1%, 5%, 10%, 20%, 30%, 40%,
50%, 60%, 70%, 80%, 90%, 100%, or greater decrease of a measure of
that parameter in a test subject relative to a reference subject.
Once an amount of change or difference in a parameter has been
correlated or associated with a disease or condition, that level
may be used in subsequent methods according to the technology
described herein. Accordingly, in some embodiments, a difference of
at least at least 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%,
90%, 100%, or more of any given structural parameter (e.g.,
tortuosity, density, volume, or any other individual structural
feature or distribution of structures or structural features as
described herein) within a data bin relative to a reference value
may be used as a threshold for methods of the technology described
herein. It should be appreciated that higher or lower or
intermediate values may be used. It also should be appreciated that
different parameters may have different threshold or reference
levels. Also, different parameters (and/or different levels for
each parameter) may be associated with different conditions or
diseases. Accordingly, specific disease or condition values or
thresholds may be identified for different parameters or
combinations thereof. These threshold values may be used for
disease detection, diagnosis, monitoring, or for any other
therapeutic, clinical, or research application described herein
(e.g., in automated methods described herein).
[0324] Accordingly, aspects of the technology described herein
provide methods and devices for obtaining and/or analyzing data
relating to internal tubular structures in casts and/or in human
and/or other animal bodies. In some embodiments, methods of the
technology described herein involve analyzing one or more
parameters (or parameter changes over time) for binned blood
vessels that have been categorized based on their size. For
example, blood vessels may be binned according to the following
non-limiting diameter ranges: about 0-10 microns, about 10-25
microns, about 25-50 microns, about 50-75 microns, about 75-100
microns, about 100-150 microns, about 150-200 microns, about
200-300 microns, about 300-400 microns, about 400-500 microns,
about 500-1,000 microns, or any combination thereof. However, any
other suitable bin size ranges (including larger, smaller, or
intermediate size ranges) may be used. In some embodiments, the
number of different bins may be between about 2 and about 10.
However, higher numbers of bins also may be used. In some
embodiments, only 2 to 5 bins are used (e.g., 2, 3, 4, or 5). In
certain embodiments, three blood vessel bin sizes are used: small,
medium, and large. In some embodiments, a single bin is chosen
having a predetermined size range and no other size ranges are
analyzed.
[0325] Profiles may be extracted from the distribution of
quantitative values for one or more structural features as
described herein (including for example, features observed in
vascular casts). In some embodiments, volume independent or density
independent profiles may be extracted from distributions by
comparing ranges within each distribution being analyzed (e.g., a
subpopulation within a single range as a percentage of the total
population across all ranges, or a ratio of subpopulations within a
first and a second range that each represent different subsets the
entire range of values).
[0326] Aspects of the technology described herein may include the
analysis of one or more regions of interest in animal disease
models (e.g., in situ and/or in casts of one or more regions of
interest). Animal disease models may be, but are not limited to,
engineered (e.g., recombinant) animals, transgenic animals,
metastatic cancer models, xenograft models, orthotopic transplant
models, etc., or any combination thereof. In some embodiments,
different animal models may have different known genetic markers
(e.g., particular mutations) associated with a disease of interest
(e.g., a cancer). Any suitable animal may be used as an animal
model, including, but not limited to, a mouse, rat, hamster, guinea
pig, pig, dog, cat, rabbit, zebrafish, or other suitable animal. It
should be appreciated that whole experimental animals may be
analyzed. However, in some embodiments, tissues and/or organs may
be analyzed. In some embodiments, models may be based on xenografts
(e.g., xenografts of cancer or tumor cells that will form cancer or
tumor tissues in a host animal). For example, human cells may be
introduced into a non-human host animal. Other uses of xenografts
include analyzing responses to certain tissue and/or organ
transplantation (e.g., a non-human tissue or organ into a human
host). In some embodiments, vascular casts of regions of interest
in an animal model may be obtained to thoroughly analyze the
vascular structures, and/or changes therein, associated with the
condition being modeled. In some embodiments, observations made on
casts may be compared (e.g., using appropriate statistical
techniques) to in vivo (e.g., in situ) observations to identify one
or more common structural characteristics and/or changes that are
statistically significant in vivo in association with a disease,
condition, or response of interest. These can then be used in
subsequent applications as described herein.
[0327] According to aspects of the technology described herein,
compounds and therapies can be evaluated in the context of an
in-vivo model such as an animal disease model. For example, a mouse
with cancer or atherosclerosis can be used to evaluate, optimize,
and identify useful therapies. Other animal models also can be
used. Aspects of the technology described herein may be useful for
high-throughput analyses because they can detect small changes in
vasculature and can be used to evaluate a therapy in a short time
period with minimal manipulation since little or no invasive
procedures are required.
[0328] Vascular analysis aspects of the technology described herein
can be used on an orthotopic model to test, for example, the
effectiveness of a drug in a short period of time. For example, the
effect of a candidate drug on angiogenesis in an orthotopic mouse
tumor model may be quantifiable after about 5 days (e.g., between 1
and 10 days, depending on the model and the drug). In contrast, a
subcutaneous cancer animal model requires approximately one month
for tumor growth to be analyzed and compared to controls.
[0329] An orthotopic model can be used to model different diseases
or clinical conditions. Examples include, cancer, tissue
regeneration, wound healing (including healing after traumatic
injury, healing after surgical intervention, healing of burnt
tissue such as skin), tissue or organ transplant therapy, medical
device implant therapy, other conditions associated with
neovascularization or changes in normal vascular structure, or any
combination of two or more of the above. However, the technology
described herein is not limited by the type of orthotopic model or
the type of disease or clinical condition that is being
analyzed.
[0330] A single orthotopic disease model animal may be useful for
testing more than one candidate drug molecule since the analysis
does not involve sacrificing the model animal. Accordingly, once a
test with a first candidate is complete, a subsequent candidate can
be evaluated in the same model animal. A series of candidates can
be tested in a single model animal, with appropriate controls,
provided the model retains features of neovascularization that are
necessary for the assay.
[0331] It should be appreciated that any of the geometrical,
structural, and/or distributional parameters described herein may
be used as biomarkers. Biomarkers of the technology described
herein can be qualified and/or quantified and compared using
standard statistical methods. These biomarkers can be compared on
individual basis, but also in combination as a signature of
vascular morphology and function. Whole signatures can be compared
between treated and untreated samples, or samples with
physiological and pathological vascular pattern.
[0332] It should be appreciated that in some embodiments, one or
more of the biomarkers described herein may be used to aid in the
diagnosis, prognosis, prediction, or other medical application
along with other types of physiological and or biological markers
(e.g., physiological measurements, genetic markers, etc., or any
combinations thereof).
[0333] It should be appreciated that aspects of the technology
described herein may be applied to features of vascular geometry
(e.g., curvature, tortuosity, distributions of vascular structural
features, etc., or any combination thereof) that are obtained from
an analysis of vascular casts (e.g., using any suitable image
analysis technique described herein or known in the art). In some
aspects, vascular casts are analyzed to identify distributions of
one or more blood vessel structural features (including, for
example, abnormal excess or absence of blood vessels or blood
vessel structures) that are associated with a disease or other
condition of interest. Structural features identified in casts may
be used as biomarkers or references to evaluate in situ
vasculature, for example, to detect indicia of a disease or other
condition of interest in a subject. Structural characteristics of
vascular casts also may be used to evaluate therapeutic treatments,
screen candidate compounds, and for other applications as described
in more detail herein. In some embodiments, one or more structural
parameters are analyzed over time (e.g., using a series of vascular
casts obtained at different time points) to monitor and/or identify
structural changes that occur during development, disease
progression or regression, or in response to therapy. In some
embodiments, structural analysis is performed on vascular casts
obtained from experimental models (e.g., whole animal models, or
organ or tissue models). However, in some embodiments, vascular
casts are obtained and analyzed for one or more regions of interest
(e.g., diseased regions) in dead animals, including for example
dead humans (e.g., human cadavers).
[0334] As used herein, a vascular cast refers to a physical
structure that is generated to represent blood vessels of an entire
vasculature or portion thereof. A cast may be obtained by perfusing
a vasculature or a vascular region (e.g., the blood vessels of an
organ, for example, of a kidney or liver) with a casting material
that solidifies (e.g., polymerizes) to form a stable structure. The
surrounding tissue and cells (e.g., including the blood vessel
walls) may be removed to reveal the cast. The cast retains the
structural features of the original blood vessels. Cast may include
structures of blood vessels of different sizes as described herein.
Certain casts are more flexible than others, certain casts are more
brittle than others. Vascular casts can be used to identify
vascular structural features with high resolution and/or to
identify correlations between structural features and conditions of
interest with high degrees of confidence since the structures of
the blood vessels are retained in the casts and other biological
structures that could interfere with an analysis are removed.
Vascular casts may be obtained using any suitable casting material.
In some embodiments, the casting agent may be a polymer. In some
embodiments, the casting agent may react with the blood vessel
walls. Non-limiting examples of casting agents include, but are not
limited to Microfil.RTM., methyl methacrylate, prepolymerized
methyl methacrylate (Mercox.TM.), Mercox.TM. CL-2B, other acrylic
resins, silicon, gold nanoparticles, Batson No. 17,
polyurethane-based casting agents (e.g., PU4ii), etc., or
combinations of two or more thereof.
[0335] It should be appreciated that casting agents may be
supplemented with contrast agents and/or other detectable agents.
Examples of contrast agents include, but are not limited to,
BaSo.sub.4 and UAc (e.g., mixed into the casting material). In some
embodiments, already polymerized casts can be soaked in OSO.sub.4
to achieve better contrast using CT imaging. In certain
embodiments, any suitable heavy metal can be mixed into the resin
to make it more radioopaque.
[0336] In some embodiments, a large volume of an animal body (e.g.,
the entire body) may be perfused with a casting agent composition.
In certain embodiments, a small volume of an animal (e.g., a
tissue, an organ or a region of either one thereof) may be perfused
with a casting agent composition. In some embodiments, a casting
agent may be perfused into a tissue or an organ or a region of
either one thereof after removal from an animal (e.g., after biopsy
or other surgical excision). In some embodiments, a casting agent
composition may be perfused into a live animal. It should be
appreciated that an animal may be sacrificed after perfusion with a
casting agent depending, in part, on the amount and type of casting
agent composition that is used and the tissue or organ to which the
casting agent composition is targeted. According to aspects of the
technology described herein, casting agent(s) may be used to
preserve in vivo structures for detailed analysis. In some
embodiments, this analysis identifies particular structural or
distribution properties that can be subsequently used as markers
for in vivo diagnostic, therapeutic, research, and/or other
applications in live animals (including humans).
[0337] In some aspects, vascular structures may be analyzed in situ
in an animal after perfusion with a casting agent composition. In
some aspects, a tissue or an organ or a region of either one
thereof may be removed from an animal for analysis (e.g., before or
after perfusion with a casting agent composition).
[0338] Accordingly, aspects of the technology described herein can
be used to represent and/or visualize blood vessels with a casting
agent or medium.
[0339] Data relating to one or more selected structures (e.g.,
structural patterns obtained from an analysis of a vascular cast)
may be obtained and/or analyzed to glean information about a
physiological condition of an animal based on the structure (or
changes in the structure). For example, patterns identified in
casts may be used as biomarkers to screen in situ vasculatures for
the presence of one or more similar patterns or to quantify the
extent of the pattern in situ. This information may be used for
diagnostic, predictive, prognostic, therapeutic, interventional,
research and/or development purposes, as well as for grading and/or
staging a disease. In some embodiments, methods of the technology
described herein may involve analyzing one or more structural
parameters (or one or more structural parameter changes over time)
based on binned structure data or information obtained for casts
(e.g., vascular casts) or in situ structures (e.g., in vivo blood
vessels).
[0340] In some embodiments, one or more structures and/or
structural changes that are identified using casts may be detected
or monitored in vivo to determine whether a predetermined disease,
condition, or response is present in vivo.
[0341] In some embodiments, structural parameters and/or structural
changes observed for vascular casts from experimental animals (or
organs or tissues) can be used as references when analyzing
vasculature in vivo. For example, structural vasculature parameters
and/or changes that are identified in casts using experimental
animal models subsequently can be detected or monitored in vivo
(e.g., in a human subject) and used to evaluate the development of
a disease, a drug response or other biological or disease property
associated with the vasculature parameters and/or changes in a
subject. In some embodiments, structural characteristics identified
in vascular casts may be used to identify one or more patient
subpopulations that are (or are predicted to be) more responsive to
a particular treatment. For example, responsive subjects may be
identified as those having one or more blood vessel characteristics
that were associated with responsiveness in animal models and
identified by analyzing vascular casts from the responsive
animals.
[0342] One or more of the characteristics described herein, or
combinations of characteristics, or related structural changes over
time, may be identified as structural patterns that can be
associated with one or more conditions of interest. Once
identified, these patterns can be used as biomarkers to identify or
monitor the conditions of interest in vivo in a subject, for
example, by analyzing the in situ vasculature of the subject (or a
portion thereof) and detecting the presence of and/or quantifying
the extent of a specific vascular structural pattern.
[0343] Accordingly, one or more of the following non-limiting
structural characteristics (e.g., combinations of 2, 3, 4, 5, 6, 7,
8, 9, 10 or all of the following structural characteristics) may be
evaluated (e.g., quantified) in vascular casts and/or in situ
(e.g., in vivo): diameter binned vessel distribution, mean vessel
diameter distribution, branching point density, vessel branching
distribution, angle of vessel branching distribution,
interbranching distances, vessel density, vessel tortuosity,
intervessel distances, luminal vessel surface, vessel dilation
(changes in vessel diameter over a segment), sinosoidalation
(dilation in sinosoids), or permeability (vessel leakiness).
Distributions of the quantified characteristics may be prepared and
analyzed (e.g., compared). However, it should be appreciated that
other structural characteristics, for example, other
characteristics described herein also may be analyzed by analyzing
and comparing distributions of those characteristics or
features.
[0344] For example, the quantification of any of the following
non-limiting features may be performed and related distributions
may be analyzed as described herein: Total Intra-Vascular Volume
(TIVV)--e.g., over the entire Tumor Vascular Tree and Region of
Interest (ROI), over only the Small Vessels Volume within the Total
Volume (or the ROI), over only the Medium Vessels Volume within the
Total Volume (or the ROI), or over only Large Vessels Volume within
Total Volume (or the ROI); Intra-Vascular Volume Distribution
(IVVD)--e.g., broken by Total Volume, Small, Mid & Large
Vessels Volumes, color encoded into small, mid, large vessels on a
segmented vascular tree (e.g., based on a Poker Chip
representation), linked vascular volume values through color
encoding of regions within a segmented vascular tree (e.g., on a
Poker Chip representation), or detected locations/regions of Max
Volume, Mid Volume, Min Volume and link to regions within a
segmented vascular tree (e.g., based on a Poker Chip
representation); Inter-Vessel Distance (IVD)--e.g., in the form of
average/Min/Max values, histograms, values in select locations (for
example single locations), color encoded Vessel Tree/ROI(s) with
IVD values & IVD Value Clusters; Inter-Branching Distance
(IBD)--e.g., in the form of average/Min/Max values, histograms,
values in select locations (for example single locations), color
encoded Vessel Tree/ROI(s) with IBD values and IBD Value Clusters;
Vascular Diameter Variability (VDV) along the length of the
vessel--e.g., in the form of histograms for the entire vascular
tree or w/in a ROI, with the ability to view such variability for a
single vessel or a group of vessels on the whole tree of within
select (ROI)s, or color encoded segments within a tree/ROI (e.g.,
based on a Poker Chip representation) based on VDV values; Vessel
Branch Curvature (VBC) and Tortuosity (VBT)--e.g., in the form of
histograms of each BC and BT for the entire vascular tree or within
select ROI(s), with the ability to view such variability for a
single vessel or a group of vessels on the whole tree or within
select ROI(s), or color encoded regions within a vascular tree/ROI
(e.g., color encoded chips a Poker Chip representation) based on BC
or BT values; or any combination of two or more thereof.
Distributions of one or more of these characteristics, or
combinations of characteristics, or related structural changes over
time, may be identified as structural patterns that can be
associated with one or more conditions of interest.
[0345] Blood vessels may be binned according to about any of the
following non-limiting diameter ranges (in microns): 0-10, 10-25,
25-50, 50-75, 75-100, 100-150, 150-200, 200-300, 300-400, 400-500,
500-1,000, or any combination thereof. However, any other suitable
bin size ranges (including larger, smaller, or intermediate) may be
used. In some embodiments, the number of different bins may be
between about 2 and about 10. However, higher numbers of bins also
may be used. In some embodiments, only 2 to 5 bins are used (e.g.,
2, 3, 4, or 5). For example, three blood vessel bin sizes may be
used: small, medium, and large diameters (e.g., small at less than
about 35 microns or about 20-35 microns, medium about 35-70 or
about 35-100 microns, and large above about 100 microns or about
100-200 microns). However, other vessel size ranges may be used to
calculate population percentages or ratios as described herein. In
some embodiments, a single bin is chosen with a predetermined size
range and no other sizes are analyzed.
[0346] In some embodiments, a parameter may be evaluated as a
percentage of the total population of vessels. For example, the
percentage of blood vessels having a particular diameter (e.g.,
20-40 microns) as a percentage of the total population of blood
vessels may be used. In some embodiments, a parameter may be
evaluated as a ratio of two subpopulations within a population of
vessels. It should be appreciated that the percentage populations
of vessels having different properties may be evaluated by
determining the relative lengths of blood vessels having different
properties within a region being analyzed. However, other
techniques may be used.
[0347] Aspects of the technology described herein relate to
business methods that may involve the marketing and/or licensing of
biomarkers associated with particular biological processes,
conditions, and/or diseases. In some embodiments, patterns (e.g.,
geometric features) of blood vessels (e.g., observed in vivo or in
casts) are analyzed to identify or evaluate associations or
correlations with certain biological processes, conditions, and/or
diseases of interest. Pattern parameters may be identified that can
be used as structural biomarkers (e.g., for clinical, diagnostic,
therapeutic, and/or research applications as described herein).
These biomarkers may be used to reduce the cost and increase the
efficiency and sensitivity of medical and research techniques. In
one embodiment, one or more biomarkers or methods of using the
biomarkers may be marketed to medical or research customers or
potential customers. In one embodiment, a fee-based service may be
provided to medical or research organizations wherein information
relating to a medical image is obtained and analyzed for the
presence of one or more biomarkers and the resulting information is
returned in exchange for a fee. The amount of the fee may be
determined, at least in part, by the type of image information that
is provided, the type and degree of analysis that is requested, and
the format and timing of the analysis. It should be understood that
aspects of the technology described herein may be applicable to
image information obtained from one or more of many different
scanning modalities (including, but not limited to, micro CT, MDCT,
rotational angiography, MRI, PACS). This information may be
received from many different sources, including, but not limited to
one or more of the following: medical centers, large pharmaceutical
companies (e.g., in association with pre-clinical evaluations or
during clinical trials), CROs (for both pre-clinical and clinical
analyses), medical laboratories and practices (e.g., scanning
centers), hospitals, clinics, medical centers, small biotechnology
companies (e.g., in association with pre-clinical evaluations or
during clinical trials), and bio-medical research organizations.
The results of the analysis then may be returned to any one of
these organizations. In some embodiments, the analysis results may
be returned to the same entity that sent the image information. In
other embodiments, the results may be returned to a different
entity (e.g., the image information may be received from a scanning
laboratory and the analysis may be returned to a physician). One or
more steps involved with receiving the information, analyzing the
structural features, processing the results and forwarding the
results to a recipient may be automated. It also should be
appreciated that one or more of these steps may be performed
outside the United States of America. Business procedures (e.g.,
marketing, selling, licensing) may be performed individually or
collaboratively.
[0348] Aspects of the technology described herein may be described
herein in the context of individual analytical steps, particular
structural features, etc. However, it should be appreciated that
any of the methods and devices described herein also may be
incorporated into a business method associated with the use of a
biomarker based on one or more blood vessel structural features or
patterns (e.g., structural features or changes observed in vascular
casts obtained from therapeutic and/or disease models or
conditions).
[0349] Aspects of the technology described herein may be automated
(e.g., using one or more computer-implemented acts described
herein). It should be appreciated that one or more pattern
parameters (e.g., individual blood vessel structural feature(s),
distributions of blood vessels or blood vessel structural features,
or combinations thereof) may be analyzed using one or more
quantitative and/or qualitative methods (e.g., based on binned
data). In some embodiments, one or more parameters may be measured
and quantified and the measurements may be analyzed using standard
quantitative and/or statistical techniques for evaluation and/or
comparison with threshold or reference values as described herein.
In certain embodiments, one or more parameters may be evaluated
using a predetermined scoring method, for example based on
predetermined factors (e.g., for binned data). Geometrical
parameters may be represented using vectors. For example, a
distribution of blood vessels, blood vessel curvatures, blood
vessel tortuosity, or blood vessel directions within a volume of
interest may be represented using a plurality of vectors. Separate
vectors may be used to represent separate vessels (e.g., vessels
for which a connectivity has not been determined during the
analysis). However, separate vectors also may be used to represent
individual segments or fragments of a single blood vessel or
portion of a vascular tree (e.g., for which connectivity has been
or may be determined during the analysis). Vasculature pattern
parameters may be analyzed using any appropriate technique for
separating and/or categorizing numerical values or scores.
[0350] In some embodiments, a score may be obtained to relate a
pattern parameter to the probability of a physiological condition
such as a disease or disease stage. Aspects of the technology
described herein can be used for in situ diagnostic, interventional
and therapeutic analysis of one or more disease loci associated
with aberrant internal structures. As used herein "in situ" means
in an animal (e.g., a human) body as opposed to in a biopsy or
other tissue sample. Aspects of the technology described herein can
be used to research structural changes associated with a disease,
for developing and evaluating disease treatments including
therapeutic drugs, and for other purposes. Aspects of the
technology described herein include automatically analyzing a
structural feature or pattern and automatically generating a score
based on the analysis.
[0351] In some embodiments, aspects of the technology described
herein include detecting and/or analyzing selected internal tubular
networks in situ in animals and/or in vascular casts. As used
herein, an internal tubular network means a network of connected
cylindrical internal body structures. Tubular networks include, but
are not limited to, cardio-vascular, respiratory,
gastro-intestinal, and genito-urinary systems and portions thereof
within animal bodies. Accordingly, the cylindrical structures may
include branched, straight, curved, and/or twisted cylindrical
elements. The cylindrical structures and elements may include not
only cylinders, but also may include flattened or otherwise
distorted regions. The cross-section of a cylindrical structure or
element may be circular, oval, approximately circular,
approximately oval, or more irregular in nature. The internal
diameter of the cylindrical elements may vary or may be
approximately the same over the region of interest. A tubular
network such as a circulatory network may be closed off from the
environment outside the animal. In contrast, tubular networks such
as respiratory and gastro-intestinal networks may be open to the
outside environment. In some embodiments, appropriate casting
and/or contrast agents (e.g., inhaled agents) may be used to
analyze respiratory and/or gastro-intestinal networks.
[0352] In one embodiment, aspects of the technology described
herein include analyzing a representation of a tubular network
(e.g., a mathematical representation of a vascular network). In one
embodiment, a representation of a network, or a portion thereof,
may be obtained (e.g., from an existing database or a remote site)
and analyzed. In another embodiment, a representation of a network,
or a portion thereof, may be generated from structural data and
then analyzed. According to aspects of the technology described
herein, an analysis may include detecting the presence or absence
of one or more structural features or patterns, measuring or
evaluating the extent of one or more structural features or
patterns, or a combination thereof.
[0353] In one embodiment, aspects of the technology described
herein are useful for selectively detecting and/or analyzing
patterns (e.g., structures) of an animal's vasculature to detect or
monitor one or more blood vessel patterns (e.g., structures) that
may be indicative of a physiological condition of the animal. A
structural pattern or feature may be detected and/or analyzed for
blood vessels of any size including, but not limited to, arteries,
arterioles, veins, venules, and capillaries.
[0354] In one embodiment, aspects of the technology described
herein are useful for selectively detecting and/or analyzing
structural features or patterns of an animal's vasculature to
detect or monitor one or more blood vessel structures that are
characteristic of disease (e.g., a disease associated with
angiogenesis). A blood vessel structure or pattern characteristic
of a disease (e.g., a disease associated with angiogenesis) may
provide an early diagnostic indication of the presence of the,
which can allow for early treatment that can improve a patient's
prognosis. In other embodiments, a blood vessel structure or
pattern characteristic of a disease (e.g., a disease associated
with angiogenesis) can be used as a marker (e.g., a biomarker) for
staging and/or grading, to monitor disease progression, evaluate a
prescribed therapy, and/or identify and/or validate a drug or
treatment regimen for the disease. Diseases associated with
abnormal vasculature structures or patterns include, but are not
limited to, cancer, cardiovascular, dermatologic (skin), arthritic,
musculoskeletal, central nervous system, neurologic, pulmonary,
renal, gastrointestinal, gynecologic, genitourinary, inflammatory,
infectious, and immunologic diseases.
[0355] A cancer may be a solid tumor or a leukemia. When the cancer
is a leukemia, methods of the technology described herein may be
directed to detecting and/or analyzing vasculature pattern(s) in
the bone marrow of an animal (e.g., human).
[0356] It also should be appreciated that aspects of the technology
described herein may include performing any combination of two or
more acts described herein and that certain acts may be omitted in
some embodiments. In one embodiment, the presence of one or more
structural abnormalities may be identified or detected in a body
region without generating and/or analyzing a structural
representation of that body region. For example, the presence of a
blood vessel abnormality may be detected directly from structure
data for a body region without generating a structural
representation of the vasculature for that entire body region. In
another embodiment, an analysis may involve selectively
representing one or more abnormal structures if they are present in
a body region without representing normal structures in that body
region (e.g., abnormal blood vessel structures may be represented
without representing any normal blood vessels, or without
representing all the normal blood vessels, without representing
most of the normal blood vessels, etc.). In another embodiment, an
abnormal vascular structure may be identified or detected without
obtaining a detailed representation of the all the blood vessels in
a body region. It may be sufficient to detect the presence of or
outline of a vascular tree in a body region and perform an analysis
that identifies or detects abnormal structures on specific blood
vessels or the presence of excessive vascularization (e.g., a clump
of neovasculature representing malignancy) without representing all
the normal details of the vascular tree or even detecting
individual blood vessels in the vascular tree. Accordingly, in some
aspects a low resolution data set for a body region may be
sufficient to detect or identify certain structural indicia of a
disease such as cancer.
[0357] Aspects of the technology described herein may include
automating one or more acts. For example, an analysis may be
automated in order to generate an output automatically. Acts of the
technology described herein may be automate using, for example, a
computer system.
[0358] As should be appreciated from the foregoing, in one
embodiment, raw or processed structure data may be obtained at a
medical or research center and sent to a computer at a remote site
where one or more of the analytical steps described above may be
performed (e.g., for a fee). The output from the analysis may be
then returned to the medical or research center either in computer
readable form to a computer at the medical or research center, in a
hard copy, in another tangible form, or in any other suitable form
including those described herein.
[0359] In another embodiment, one or more software programs that
implement one or more functionalities described herein may be
provided and installed at a medical or research center (e.g., for a
fee). The programs can be provided on disk, downloaded from an
internal or remote (e.g., external) site, or loaded in any suitable
manner. Reference information that is used in any functionality
described herein may be provided along with the software or
separately. In one embodiment, reference information (e.g.,
information relating to normal or abnormal blood vessel structures)
may be available on disk, downloaded from an internal or remote
(e.g., external) site, or loaded in any suitable manner.
[0360] As used herein, "remote" means at a site that is different
from the immediate location of the imaging device (e.g., the
medical scanner). The remote site can be a central computer or
computing facility at a hospital, medical, or research center
(e.g., within the network or intranet of the center), or can be
outside the hospital, medical, or research center (e.g., outside
the network or intranet of the center). The remote site can be in
the same state, in a different state, or in a different country
from the site of data acquisition by the imaging device.
[0361] In some embodiments, multimodal analyses (e.g., using
structure data from two or more different types of imaging devices)
may be used together. Accordingly, aspects of the present
technology described herein may include the ability to process and
analyze different types of structure data and either combine the
results to generate a combined output, or to generate a separate
output is generated for each imaging modality. In some embodiments,
an organ, tissue, or animal perfused with a casting agent and/or an
imaging agent may be sent to an imaging center for analysis.
[0362] In some embodiments, in vivo and/or ex vivo casting methods
of the technology described herein can be used to identify one or
more vascular patterns (e.g., including one or more structural
parameters, structure distributions, combinations thereof) and/or
time-dependent changes thereof that can be used as biomarker(s) for
a disease or a response to a therapy, or for monitoring patients
for indicia of disease or response to therapy, or for other
applications where vascular information may be informative.
Accordingly, such vascular patterns or changes thereof identified
according to methods of the technology described herein can be used
for diagnostic, interventional, therapeutic, research, and
treatment development and evaluation. Non-limiting examples of some
of these embodiments are described below.
EXAMPLES
Example 1
Xenotopic Tumor Models
[0363] A tumor model can be generated by inoculating human
non-small cell lung tumor cell line (A549 from ATCC, Inc.)
subcutaneously in immunodeficient mice (SCID). SCID male mice (6-8
weeks old from Charles River Inc.) are inoculated subcutaneously in
the lower back with a suspension of 1.times.10.sup.6 human lung
tumor cells (A549) in 0.2 ml of PBS. All mice are fed normal chow
diet throughout the duration of the experiment. All mice weights
are measured throughout the experiment. Tumor size is measured with
calipers twice-a-week and tumor volume is calculated using the
formula Length.sup.2.times.Width.times.0.52. All mice are
randomized into two treatment groups (approximately 10 mice per
group) when the median tumor volume reaches approximately 500
mm.sup.3. The treatment groups can be treated according to the
following schedule using intraperitoneal (i.p.) administration of
either a control composition or an anti-angiogenic compound. For
example, different levels of an anti-angiogenic compound can be
used and the results compared to a control group that is not
treated with an anti-angiogenic compound (e.g., Avastin.RTM.
available from Genentech, South San Francisco, Calif.). For
example:
[0364] Group 1: Control group--treated with saline/PBS twice a
week.
[0365] Group 2: High Avastin.RTM.--treated with Avastin.RTM. at 5
mg/kg/i.p. twice a week.
[0366] Group 3: Low Avastin.RTM.--treated with Avastin.RTM. at 0.5
mg/kg/i.p. twice a week.
[0367] Experiments are terminated 1.5 weeks after initial
treatment.
[0368] At the end-point, all mice are anesthetized and systemically
perfused with a casting agent.
Example 2
Perfusion with Casting Agent
[0369] Perfusion with a casting agent, Mercox (available from Ladd
Research, Williston, Vt.) can be performed as follows. An initial
anticoagulation step for each animal is performed using an i.v.
injection of heparin (10,000 U/ml, 0.3 cc/mouse). After 30 minutes,
the animals are anesthetized. Each animal's heart is cannulated and
the animal perfused with warm physiological saline at physiological
pressure (with an open vein draining the organ or with an open vena
cava). Perfusion is continued until the organ or animal is clear of
blood. Mercox monomer is filtered through a 0.5 .mu.m filter and a
casting resin is prepared by mixing 8 ml Mercox, 2 ml
methylmethacrylate, and 0.3 ml catalyst. The resin is infused
through the same cannula until the onset of polymerization (the
resin changes color to brown and emits heat, .about.10 min). The
organ or animal is carefully immersed in a 60.degree. C. water bath
for 2 hours (or overnight in a sealed container). The tissue is
removed by incubating in alternating rinses of 5% KOH and distilled
water (for example in a 60.degree. C. water bath sealed) followed
by thorough rinsing in distilled water. The cast is cleaned in 5%
formic acid for 15 minutes and rinsed thoroughly in distilled water
and frozen in distilled water. The resulting block of ice is
lyophilized (care should be taken not to melt the ice, the ice
should melt as it lyophilizes). The resulting cast can be analyzed
to identify one or more structural characteristics of interest.
Example 3
Xenotopic Tumor Models Response to Anti-Angiogenic Therapy
[0370] Xenotopic mouse models obtained as described in Example 1
were treated with either a control solution of saline/PBS or an
anti-angiogenic preparation of Avastin.RTM. at 0.5 mg/kg/i.p. as
described above. At the end-point, vascular casts were prepared as
described in Example 2 above and analyzed for two treated mice
(both treated with Avastin.RTM. at 0.5 mg/kg/i.p.) and one control
mouse. The resulting vascular casts were scanned using a micro
CT-scanner and the results of the structural analysis are shown in
FIGS. 14-17. The analysis was performed by determining the number
of blood vessels within bins of different diameter ranges for the
xenotopic tumor in the treated and control animals. The bins were
each 13.8 .mu.m wide and the smallest bin included blood vessels
having a diameter of between 20.7 .mu.m and 34.5 .mu.m. Mean tumor
volumes did not differ significantly between the groups at the end
of the experiment. However differences in blood vessel diameter
distributions were detected as shown in FIGS. 14-17. FIG. 14 shows
the resulting vessel population distribution. Treated tumors had
20% less small diameter sized vessels than untreated tumors, and
treated tumors had a higher percentage of middle diameter sized
vessels than untreated tumors. The blood vessel population
distributions were consistent for both treated animals. FIG. 15
shows the vessel population ratio between small (approximately
21-35 .mu.m) and middle (approximately 35-49 .mu.m) size vessels in
the tumors of the control and treated animals. The ratio decreased
after inhibitor treatment with Avastin.RTM., and this ratio was
consistent within the treated group. FIG. 16 shows the vessel
population ratio between large (approximately 147-161 .mu.m) and
middle (approximately 33-77 .mu.m) size vessels. The ratio
decreased after treatment with Avastin.RTM., and this ratio was
consistent within the treated group. Additional experimental
results are shown in FIGS. 17-19.
[0371] The following considerations apply to the specific examples
and the entire written specification herein (including the summary,
detailed description, and claims). It should be appreciated that
casts, like in situ blood vessels, are three-dimensional
structures. Accordingly, imaging and analytical techniques
described herein provide information about three-dimensional
structural characteristics. In some embodiments, techniques are
used to generate three-dimensional representations of vascular
casts and/or in situ blood vessels. In some embodiments, techniques
are used to generate three-dimensional images of vascular casts
and/or in situ blood vessels. The three-dimensional representations
and/or images can be analyzed as described herein.
[0372] However, it should be appreciated that aspects of the
technology described herein are not limited to three-dimensional
structural characteristics. In some embodiments, aspects of
vascular casts and/or in situ blood vessels may be represented
and/or imaged in one or two dimensions and an analysis of one or
two-dimensional features may be performed and used as described
herein. It also should be appreciated that the structural features
described herein may be measured or quantified using any
appropriate units, including numbers, lengths or distances, angles,
percentages, etc., or any combination thereof, further including
any of these units as a function of volume or area. Similarly, it
should be appreciated that vascular changes over time or in
response to treatment may involve an increase or a decrease of one
or more of these structural features. For example, an increase in
structures associated with angiogenesis may be associated with
certain disease progressions. In contrast, a decrease in structures
associated with angiogenesis may be associated with disease
regression (e.g., in response to treatment).
[0373] It also should be appreciated that descriptions herein
related to obtaining distributions of quantitative values for
vessel parameters within a region of interest are preferably based
on methodologies that detect and quantify all or substantially all
of the detectable vessels within the region of interest based on
the detection technique that is used for that analysis. Different
techniques may have different efficiencies. However, profiles and
comparisons are preferably based on data from the same or
equivalent detection and/or reconstruction techniques. It also
should be appreciated that comparisons and/or analyses described
herein may involve a statistical analysis using one or more
standard statistical techniques to determine whether a change in a
structure or pattern or other characteristic described herein
(e.g., an increase or decrease over time, or in response to a
therapeutic drug), or a difference or similarity between two
structures or patterns or other characteristics described herein
are statistically significant.
[0374] Having thus described several aspects of at least one
embodiment of this technology described herein, it is to be
appreciated various alterations, modifications, and improvements
will readily occur to those skilled in the art. Such alterations,
modifications, and improvements are intended to be within the
spirit and scope of the technology described herein. Any suitable
analytical techniques may be used for perfused tissue and organs
according to the methods described herein, including for example,
the analytical techniques that are described in PCT US2005/047081
and PCT US2007/026048 the disclosures of which are incorporated
herein by reference in their entirety. Accordingly, the foregoing
description and embodiments are by way of example only. In the
event of conflict between different disclosures, the disclosure of
the present application shall control.
[0375] It should be appreciated from the foregoing, there are
numerous aspects of the technology described herein described
herein that can be used independently of one another or in any
combination. In particular, any of the herein described operations
may be employed in any of numerous combinations and procedures. In
addition, aspects of the technology described herein can be used in
connection with a variety of types of images or any dimensionality.
Moreover, one or more automatic operations can be used in
combination with one or more manual operations, as the aspects of
the technology described herein are not limited in this respect.
Distribution analyses, however obtained, may be used to facilitate
the characterization of any of various morphological changes to
tissue and/or to assist in assessing the efficacy of treatment
using any of the herein described techniques, alone or in
combination.
[0376] An illustrative implementation of a computer system 2800
that may be used to implement one or more of the techniques
described herein (e.g., any of the processes described herein such
as processes 2000, 2100, 2200, 2300, and 2500 related to generating
a vessel network at least in part by linking vessel centerline
voxels) is shown in FIG. 28. Computer system 2800 may include one
or more processors 2810 and one or more non-transitory
computer-readable storage media (e.g., memory 2820 and one or more
non-volatile storage media 2830). The processor 2810 may control
writing data to and reading data from the memory 2820 and the
non-volatile storage device 2830 in any suitable manner, as the
aspects of the invention described herein are not limited in this
respect.
[0377] To perform functionality and/or techniques described herein,
the processor 2810 may execute one or more instructions stored in
one or more computer-readable storage media (e.g., the memory 2820,
storage media, etc.), which may serve as non-transitory
computer-readable storage media storing instructions for execution
by the processor 2810. Computer system 2800 may also include any
other processor, controller or control unit needed to route data,
perform computations, perform I/O functionality, etc. For example,
computer system 2800 may include any number and type of input
functionality to receive data and/or may include any number and
type of output functionality to provide data, and may include
control apparatus to operate any present I/O functionality.
[0378] In connection with the techniques described herein, one or
more programs configured to perform one or more processes related
to generating a vessel network (examples of which have been
provided above) and/or any other suitable processes may be stored
on one or more computer-readable storage media of computer system
2800. Processor 2810 may execute any one or combination of such
programs that are available to the processor by being stored
locally on computer system 2800 or accessible over a network. Any
other software, programs or instructions described herein may also
be stored and executed by computer system 2800. Computer 2800 may
be a standalone computer, server, part of a distributed computing
system, mobile device, etc., and may be connected to a network and
capable of accessing resources over the network and/or communicate
with one or more other computers connected to the network.
[0379] Implementation of some of the techniques described herein
(e.g., linking centerline voxels, identifying branch points, etc.)
on a computer system such as computer 2800 is an integral component
of practicing these techniques, as aspect of these techniques
cannot be realized absent computer implementation
[0380] The herein-described embodiments of the present technology
described herein can be implemented in any of numerous ways. For
example, linking of centerline voxels may be implemented using
hardware, software or a combination thereof. When implemented in
software, the software code can be executed on any suitable
processor or collection of processors, whether provided in a single
computer or distributed among multiple computers. It should be
appreciated that any component or collection of components that
perform the functions described herein can be generically
considered as one or more controllers that control the
herein-discussed functions. The one or more controllers can be
implemented in numerous ways, such as with dedicated hardware, or
with general purpose hardware (e.g., one or more processors) that
is programmed using microcode or software to perform the functions
recited herein.
[0381] The terms "program" or "software" are used herein in a
generic sense to refer to any type of computer code or set of
processor-executable instructions that can be employed to program a
computer or other processor to implement various aspects of
embodiments as discussed above. Additionally, it should be
appreciated that according to one aspect, one or more computer
programs that when executed perform methods of the disclosure
provided herein need not reside on a single computer or processor,
but may be distributed in a modular fashion among different
computers or processors to implement various aspects of the
disclosure provided herein.
[0382] Processor-executable instructions may be in many forms, such
as program modules, executed by one or more computers or other
devices. Generally, program modules include routines, programs,
objects, components, data structures, etc. that perform particular
tasks or implement particular abstract data types. Typically, the
functionality of the program modules may be combined or distributed
as desired in various embodiments.
[0383] Also, data structures may be stored in one or more
non-transitory computer-readable storage media in any suitable
form. For simplicity of illustration, data structures may be shown
to have fields that are related through location in the data
structure. Such relationships may likewise be achieved by assigning
storage for the fields with locations in a non-transitory
computer-readable medium that convey relationship between the
fields. However, any suitable mechanism may be used to establish
relationships among information in fields of a data structure,
including through the use of pointers, tags or other mechanisms
that establish relationships among data elements.
[0384] Also, various inventive concepts may be embodied as one or
more processes, of which examples (see e.g., FIGS. 20-23 and 25)
have been provided. The acts performed as part of each process may
be ordered in any suitable way. Accordingly, embodiments may be
constructed in which acts are performed in an order different than
illustrated, which may include performing some acts concurrently,
even though shown as sequential acts in illustrative
embodiments.
[0385] As used herein in the specification and in the claims, the
phrase "at least one," in reference to a list of one or more
elements, should be understood to mean at least one element
selected from any one or more of the elements in the list of
elements, but not necessarily including at least one of each and
every element specifically listed within the list of elements and
not excluding any combinations of elements in the list of elements.
This definition also allows that elements may optionally be present
other than the elements specifically identified within the list of
elements to which the phrase "at least one" refers, whether related
or unrelated to those elements specifically identified. Thus, as a
non-limiting example, "at least one of A and B" (or, equivalently,
"at least one of A or B," or, equivalently "at least one of A
and/or B") can refer, in one embodiment, to at least one,
optionally including more than one, A, with no B present (and
optionally including elements other than B); in another embodiment,
to at least one, optionally including more than one, B, with no A
present (and optionally including elements other than A); in yet
another embodiment, to at least one, optionally including more than
one, A, and at least one, optionally including more than one, B
(and optionally including other elements); etc.
[0386] The phrase "and/or," as used herein in the specification and
in the claims, should be understood to mean "either or both" of the
elements so conjoined, i.e., elements that are conjunctively
present in some cases and disjunctively present in other cases.
Multiple elements listed with "and/or" should be construed in the
same fashion, i.e., "one or more" of the elements so conjoined.
Other elements may optionally be present other than the elements
specifically identified by the "and/or" clause, whether related or
unrelated to those elements specifically identified. Thus, as a
non-limiting example, a reference to "A and/or B", when used in
conjunction with open-ended language such as "comprising" can
refer, in one embodiment, to A only (optionally including elements
other than B); in another embodiment, to B only (optionally
including elements other than A); in yet another embodiment, to
both A and B (optionally including other elements); etc.
[0387] Use of ordinal terms such as "first," "second," "third,"
etc., in the claims to modify a claim element does not by itself
connote any priority, precedence, or order of one claim element
over another or the temporal order in which acts of a method are
performed. Such terms are used merely as labels to distinguish one
claim element having a certain name from another element having a
same name (but for use of the ordinal term).
[0388] The phraseology and terminology used herein is for the
purpose of description and should not be regarded as limiting. The
use of "including," "comprising," "having," "containing",
"involving", and variations thereof, is meant to encompass the
items listed thereafter and additional items.
[0389] Having described several embodiments of the techniques
described herein in detail, various modifications, and improvements
will readily occur to those skilled in the art. Such modifications
and improvements are intended to be within the spirit and scope of
the disclosure. Accordingly, the foregoing description is by way of
example only, and is not intended as limiting. The techniques are
limited only as defined by the following claims and the equivalents
thereto.
* * * * *