U.S. patent application number 14/627713 was filed with the patent office on 2015-08-20 for method of analyzing a medical image.
The applicant listed for this patent is LOMA LINDA UNIVERSITY. Invention is credited to Stephen Ashwal, Bir Bhanu, Nirmalya Ghosh, Andre Obenaus.
Application Number | 20150235362 14/627713 |
Document ID | / |
Family ID | 44507191 |
Filed Date | 2015-08-20 |
United States Patent
Application |
20150235362 |
Kind Code |
A1 |
Ghosh; Nirmalya ; et
al. |
August 20, 2015 |
METHOD OF ANALYZING A MEDICAL IMAGE
Abstract
A method of analyzing a medical image, where the medical image
comprises one or more than one region of interest, and where the
method comprises a) providing the medical image comprising a set of
actual image values; b) rescaling the actual image values to
produce corresponding rescaled image values and to produce a
rescaled image from the rescaled image values; c) deriving a
histogram of the rescaled image values; d) using the histogram to
derive an adaptive segmentation threshold; e) using the adaptive
segmentation threshold to recursively split the rescaled image; f)
terminating the recursive splitting of the sub(sub) images using
one or more than one predetermined criteria; and g) identifying one
sub(sub) image in the terminated Hierarchical Region Splitting Tree
which comprises the region of interest.
Inventors: |
Ghosh; Nirmalya; (San
Bernardino, CA) ; Ashwal; Stephen; (Riverside,
CA) ; Obenaus; Andre; (Colton, CA) ; Bhanu;
Bir; (Riverside, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
LOMA LINDA UNIVERSITY |
Loma Linda |
CA |
US |
|
|
Family ID: |
44507191 |
Appl. No.: |
14/627713 |
Filed: |
February 20, 2015 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
14272299 |
May 7, 2014 |
8965089 |
|
|
14627713 |
|
|
|
|
13580947 |
Aug 24, 2012 |
8731261 |
|
|
PCT/US2011/025943 |
Feb 23, 2011 |
|
|
|
14272299 |
|
|
|
|
61327630 |
Apr 23, 2010 |
|
|
|
61307396 |
Feb 23, 2010 |
|
|
|
Current U.S.
Class: |
382/128 |
Current CPC
Class: |
G06K 9/4642 20130101;
G06T 2207/10081 20130101; G06T 7/11 20170101; G06T 7/162 20170101;
G06K 9/3233 20130101; G06T 2207/20036 20130101; G06T 2207/20076
20130101; G06K 2009/4666 20130101; G06T 2207/30024 20130101; G06T
2207/30052 20130101; G06T 2207/30056 20130101; G06T 2207/30061
20130101; G06T 7/0012 20130101; G06T 7/136 20170101; G06K 9/38
20130101; G06T 7/155 20170101; G06T 2207/10104 20130101; G06T
2207/10088 20130101; G06T 7/143 20170101; G06K 9/52 20130101; G06T
2207/30016 20130101; G06T 2207/30088 20130101; G06T 2207/10116
20130101; G06T 2207/20072 20130101; G06T 2207/30084 20130101; G06K
9/46 20130101; G06T 2207/30048 20130101 |
International
Class: |
G06T 7/00 20060101
G06T007/00; G06K 9/52 20060101 G06K009/52; G06K 9/46 20060101
G06K009/46 |
Goverment Interests
STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
[0002] This invention was made with United States Government
support under Cooperative Agreement Number DAMD17-97-2-7016 with
the National Medical Test-Bed, Inc., United States Army Medical
Research Acquisition Activity (USAMRAA). The United States
Government has certain rights in this invention.
Claims
1. A method of analyzing a medical image, the medical image
comprising one or more than one region of interest, the method
comprising: a) providing the medical image comprising a set of
actual image values; b) rescaling the actual image values to
produce corresponding rescaled image values and to produce a
rescaled image from the rescaled image values; c) deriving a
histogram of the rescaled image values; d) using the histogram to
derive an adaptive segmentation threshold that can be used to split
the rescaled image into two sub-images, a first sub-image with
intensities at or below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold, or a first sub-image with intensities below the adaptive
segmentation threshold and a second sub-image with intensities at
or above the adaptive segmentation threshold, or a first sub-image
with intensities below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold; e) using the adaptive segmentation threshold to
recursively split the rescaled image to generate a Hierarchical
Region Splitting Tree of sub(sub) images based on consistency of
the rescaled image values of the rescaled image; f) terminating the
recursive splitting of the sub(sub) images using one or more than
one predetermined criteria thereby completing the Hierarchical
Region Splitting Tree; and g) identifying one sub(sub) image in the
terminated Hierarchical Region Splitting Tree which comprises the
region of interest; the method further comprising performing a
secondary rescaling of the rescaled image values of every rescaled
sub(sub) image in the Hierarchical Region Splitting Tree back to
the actual image values present in the medical image to create a
secondary rescaled medical image, thereby producing a secondarily
rescaled sub(sub) image comprising the region of interest; where
the rescaled image values fit in [0,255] unsigned 8-bit integer
range; where the predetermined criteria is selected from the group
consisting of area threshold=50 pixels and (standard deviation
threshold=10 rscVals (StdDevTh=10 rscVals) and kurtosis
threshold=1.5); where the region of interest is a representation of
an abnormality in the living human tissue, and where the method
further comprises quantifying the abnormality in the living human
tissue; where the method further comprises performing a secondary
resealing of the resealed image values in every resealed sub(sub)
image in the Hierarchical Region Splitting Tree back to the actual
image values present in the medical image to create a secondary
resealed medical image, and determining an image value or a set of
image values of actual image values in the medical image after the
secondary resealing, where the image value or a set of image values
of actual image values determined identifies the abnormality
represented in the medical image for the modality being used to
generate the medical image provided; and where the method further
comprises preparing a mask of the sub(sub) image containing the
representation of the abnormality, and cleaning the mask to remove
small outlier regions to generate a cleaned mask of the sub(sub)
image containing the representation of the abnormality.
2. The method of claim 1, where the one sub(sub) image in the
terminated Hierarchical Region Splitting Tree comprising the region
of interest is two-dimensional or three-dimensional.
3. The method of claim 1, where the secondarily resealed sub(sub)
image in the terminated Hierarchical Region Splitting Tree
comprising the region of interest is two-dimensional or
three-dimensional.
4. (canceled)
5. (canceled)
6. The method of claim 1, where the medical image is selected from
the group consisting of a computed tomography scan, a magnetic
resonance image, a positron emission tomography scan and an
X-ray.
7. (canceled)
8. (canceled)
9. The method of claim 1, where the abnormality is an injury to
living human tissue and the method detects a representation of the
injury.
10. The method of claim 9, where the method qualifies the
representation of the injury.
11. The method of claim 1, where the medical image is of human
tissue selected from the group consisting of brain, heart,
intestines, joints, kidneys, liver, lungs and spleen.
12. The method of claim 1, where the medical image provided is in a
hard copy form, and where the method further comprises preparing a
digital form of the medical image before providing the medical
image.
13. The method of claim 1, where the abnormality is a genetic
malformation.
14. The method of claim 1, where determining the image value or a
set of image values of actual image values is made before or after
the step of providing the medical image.
15. (canceled)
16. The method of claim 1, where the medical image is a magnetic
resonance image and the modality being used to generate the medical
image provided is selected from the group consisting of an apparent
diffusion coefficient map, a magnetic susceptibility map and a T2
map.
17. A method of detecting a core of an injury and detecting a
penumbra of an injury in living human tissue, and distinguishing
the core from the penumbra, the method comprising: a) detecting one
sub(sub) image in the terminated Hierarchical Region Splitting Tree
comprising the region of interest according to the method of claim
1, where the abnormality is an injury; b) determining the mask of
the injury; c) determining a sub-tree below the detected injury
sub-image using the injury sub-image as the root of the sub-tree;
d) determining the soft threshold image values for separating the
core from the penumbra; e) comparing the soft threshold image
values inside the sub-tree to find the penumbra and a mask of the
penumbra; and f) determining the mask of the core by subtracting
the mask of the penumbra from the mask of the injury.
18. The method of claim 17, further comprising determining
different gradations of the core and the penumbra.
19. The method of claim 1, further comprising quantifying the
spatiotemporal evolution of an injury in living human tissue.
20. A method of detecting the effects of endogenous or implanted
stem cells on living human tissue, the method comprising: a)
determining magnetic resonance image values of labeled and
implanted stem cells; b) detecting the stem cells outside of the
region of interest using a method according to claim 1; and c)
detecting the stem cells inside of the region of interest using a
method according to claim 1.
21. The method of claim 19, further comprising quantifying
spatiotemporal activities of implanted labeled stem cells in the
living human tissue.
22. A method of analyzing a medical image, the medical image
comprising one or more than one region of interest, the method
comprising: configuring at least one processor to perform the
functions of: 1) providing the medical image comprising a set of
actual image values; 2) rescaling the actual image values to
produce corresponding rescaled image values and to produce a
rescaled image from the rescaled image values; 3) deriving a
histogram of the rescaled image values; 4) using the histogram to
derive an adaptive segmentation threshold that can be used to split
the rescaled image into two sub-images, a first sub-image with
intensities at or below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold, or a first sub-image with intensities below the adaptive
segmentation threshold and a second sub-image with intensities at
or above the adaptive segmentation threshold, or a first sub-image
with intensities below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold; 5) using the adaptive segmentation threshold to
recursively split the rescaled image to generate a Hierarchical
Region Splitting Tree of sub(sub) images based on consistency of
the rescaled image values of the rescaled image; 6) terminating the
recursive splitting of the sub(sub) images using one or more than
one predetermined criteria thereby completing the Hierarchical
Region Splitting Tree; and 7) identifying one sub(sub) image in the
terminated Hierarchical Region Splitting Tree which comprises the
region of interest; the method further comprising performing a
secondary rescaling of the rescaled image values of every rescaled
sub(sub) image in the Hierarchical Region Splitting Tree back to
the actual image values present in the medical image to create a
secondary rescaled medical image, thereby producing a secondarily
rescaled sub(sub) image comprising the region of interest; where
the rescaled image values fit in [0,255] unsigned 8-bit integer
range; where the predetermined criteria is selected from the group
consisting of area threshold=50 pixels and (standard deviation
threshold=10 rscVals (StdDevTh=10 rscVals) and kurtosis
threshold=1.5); where the region of interest is a representation of
an abnormality in the living human tissue, and where the method
further comprises quantifying the abnormality in the living human
tissue; where the method further comprises performing a secondary
rescaling of the rescaled image values in every rescaled sub(sub)
image in the Hierarchical Region Splitting Tree back to the actual
image values present in the medical image to create a secondary
rescaled medical image, and determining an image value or a set of
image values of actual image values in the medical image after the
secondary rescaling, where the image value or a set of image values
of actual image values determined identifies the abnormality
represented in the medical image for the modality being used to
generate the medical image provided; where the medical image is a
magnetic resonance image and the modality being used to generate
the medical image provided is selected from the group consisting of
an apparent diffusion coefficient map, a magnetic susceptibility
map and a T2 map; and where the method further comprises preparing
a mask of the sub(sub) image containing the representation of the
abnormality, and cleaning the mask to remove small outlier regions
to generate a cleaned mask of the sub(sub) image containing the
representation of the abnormality.
23. The method of claim 22, where the one sub(sub) image in the
terminated Hierarchical Region Splitting Tree comprising the region
of interest is two-dimensional or three-dimensional.
24. The method of claim 22, where the secondarily rescaled sub(sub)
image in the terminated Hierarchical Region Splitting Tree
comprising the region of interest is two-dimensional or
three-dimensional.
25. (canceled)
26. (canceled)
27. The method of claim 22, where the medical image is selected
from the group consisting of a computed tomography scan, a magnetic
resonance image, a positron emission tomography scan and an
X-ray.
28. (canceled)
29. (canceled)
30. (canceled)
31. (canceled)
32. (canceled)
33. (canceled)
34. (canceled)
35. (canceled)
36. (canceled)
37. (canceled)
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] The present application is a continuation of U.S. patent
application Ser. No. 14/272,299 titled, "Method of Analyzing a
Medical Image," filed May 7, 2014, which is a continuation of U.S.
patent application Ser. No. 13/580,947 titled "Method of Analyzing
a Medical Image," filed Aug. 24, 2012, which is a national stage of
International Patent Application No. PCT/US2011/025943, titled
"Method of Analyzing a Medical Image," filed Feb. 23, 2011, which
claims the benefit of U.S. Provisional Patent Application No.
61/307,396 entitled "Method for Distinguishing Normal from Abnormal
Tissue," filed Feb. 23, 2010; and U.S. Provisional Patent
Application No. 61/327,630 entitled "Method of Analyzing a Medical
Image," filed Apr. 23, 2010, the contents of which are incorporated
in this disclosure by reference in their entirety.
BACKGROUND
[0003] Medical imaging, such as for example computed tomography
scans (CT scan), magnetic resonance images (MRI), positron emission
tomography scans (PET scan) and X-rays are essential for diagnosing
and for monitoring the treatment of patients. The gold standard for
analyzing a medical image is by human visual inspection and
analysis of the image by a trained technician; however, human
visual inspection and analysis of the image usually takes from
minutes to hours to complete. Often, time is critical in reading a
medical image and supplying the results to the personnel treating
the patient.
[0004] Therefore, there is a need for a method of analyzing a
medical image that provides similar results as human visual
inspection and analysis of the image by a trained technician that
takes substantially less time than human visual inspection and
analysis of the image by a trained technician.
SUMMARY
[0005] According to one embodiment of the present invention, there
is provided a method of analyzing a medical image, where the
medical image comprises one or more than one region of interest,
and where the method comprises a) providing the medical image
comprising a set of actual image values; b) rescaling the actual
image values to produce corresponding rescaled image values and to
produce a rescaled image from the rescaled image values; c)
deriving a histogram of the rescaled image values; d) using the
histogram to derive an adaptive segmentation threshold that can be
used to split the rescaled image into two sub-images, a first
sub-image with intensities at or below the adaptive segmentation
threshold and a second sub-image with intensities above the
adaptive segmentation threshold, or a first sub-image with
intensities below the adaptive segmentation threshold and a second
sub-image with intensities at or above the adaptive segmentation
threshold, or a first sub-image with intensities below the adaptive
segmentation threshold and a second sub-image with intensities
above the adaptive segmentation threshold; e) using the adaptive
segmentation threshold to recursively split the rescaled image to
generate a Hierarchical Region Splitting Tree of sub(sub) images
based on consistency of the rescaled image values of the rescaled
image; f) terminating the recursive splitting of the sub(sub)
images using one or more than one predetermined criteria thereby
completing the Hierarchical Region Splitting Tree; and g)
identifying one sub(sub) image in the terminated Hierarchical
Region Splitting Tree which comprises the region of interest.
[0006] According to another embodiment of the present invention,
there is provided a method of analyzing a medical image, where the
medical image comprises one or more than one region of interest.
The method comprises a) configuring at least one processor to
perform the functions of: 1) providing the medical image comprising
a set of actual image values; 2) rescaling the actual image values
to produce corresponding rescaled image values and to produce a
rescaled image from the rescaled image values; 3) deriving a
histogram of the rescaled image values; 4) using the histogram to
derive an adaptive segmentation threshold that can be used to split
the rescaled image into two sub-images, a first sub-image with
intensities at or below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold, or a first sub-image with intensities below the adaptive
segmentation threshold and a second sub-image with intensities at
or above the adaptive segmentation threshold; 5) using the adaptive
segmentation threshold to recursively split the rescaled image to
generate a Hierarchical Region Splitting Tree of sub(sub) images
based on consistency of the rescaled image values of the rescaled
image; 6) terminating the recursive splitting of the sub(sub)
images using one or more than one predetermined criteria thereby
completing the Hierarchical Region Splitting Tree; and 7)
identifying one sub(sub) image in the terminated Hierarchical
Region Splitting Tree which comprises the region of interest.
[0007] In one embodiment, the method further comprises performing a
secondary rescaling of the rescaled image values of every rescaled
sub(sub) image in the Hierarchical Region Splitting Tree back to
the actual image values present in the medical image to create a
secondary rescaled medical image, thereby producing a secondarily
rescaled sub(sub) image comprising the region of interest. In
another embodiment, the one sub(sub) image in the terminated
Hierarchical Region Splitting Tree comprising the region of
interest is two-dimensional. In another embodiment, the secondarily
rescaled sub(sub) image in the terminated Hierarchical Region
Splitting Tree comprising the region of interest is
three-dimensional. In a preferred embodiment, the secondarily
rescaled one sub(sub) image in the terminated Hierarchical Region
Splitting Tree comprising the region of interest is
two-dimensional. In a preferred embodiment, the one sub(sub) image
in the terminated Hierarchical Region Splitting Tree comprising the
region of interest is three-dimensional. In one embodiment, the
medical image is selected from the group consisting of a computed
tomography scan, a magnetic resonance image, a positron emission
tomography scan and an X-ray. In one embodiment, the one sub(sub)
image in the terminated Hierarchical Region Splitting Tree
comprising the region of interest is two-dimensional. In one
embodiment, the one sub(sub) image in the terminated Hierarchical
Region Splitting Tree comprising the region of interest is
three-dimensional. In one embodiment, the region of interest is a
representation of an injury to living human tissue and the method
detects the representation of the injury. In another embodiment,
the method qualifies the representation of the injury. In one
embodiment, the medical image is of human tissue is selected from
the group consisting of brain, heart, intestines, joints, kidneys,
liver, lungs and spleen. In another embodiment, the medical image
provided is in a hard copy form, and where the method further
comprises preparing a digital form of the medical image before
providing the medical image. In one embodiment, the rescaled image
values fit in [0,255] unsigned 8-bit integer range. In another
embodiment, the predetermined criteria are selected from the group
consisting of area threshold=50 pixels and (standard deviation
threshold=10 rscVals (StdDevTh=10 rscVals) and kurtosis
threshold=1.5).
[0008] In one embodiment, there is provided a method of detecting
an abnormality in living human tissue. The method comprises
analyzing a medical image according to the present invention, where
the region of interest is a representation of the abnormality in
the living human tissue, and where the method further comprises
quantifying the abnormality in the living human tissue. In one
embodiment, the abnormality is selected from the group consisting
of a genetic malformation and an injury. In one embodiment, the
method further comprises performing a secondary rescaling of the
rescaled image values in every rescaled sub(sub) image in the
Hierarchical Region Splitting Tree back to the actual image values
present in the medical image to create a secondary rescaled medical
image; and where the method further comprises determining an image
value or a set of image values of actual image values in the
medical image after the secondary rescaling, where the image value
or a set of image values of actual image values determined
identifies the abnormality represented in the medical image for the
modality being used to generate the medical image provided. In one
embodiment, determining the image value or a set of image values of
actual image values is made before the step of providing the
medical image. In another embodiment, determining the image value
or a set of image values of actual image values is made after the
step of providing the medical image. In one embodiment, the medical
image is a magnetic resonance image and the modality being used to
generate the medical image provided is selected from the group
consisting of an apparent diffusion coefficient map, a magnetic
susceptibility map and a T2 map. In another embodiment, the method
further comprises preparing a mask of the sub(sub) image containing
the representation of the abnormality, and cleaning the mask to
remove small outlier regions to generate a cleaned mask of the
sub(sub) image containing the representation of the
abnormality.
[0009] In one embodiment, there is provided a method of detecting a
core of an injury and detecting a penumbra of an injury in living
human tissue, and distinguishing the core from the penumbra, the
method comprising: a) detecting one sub(sub) image (the injury
sub-image) in the terminated Hierarchical Region Splitting Tree
comprising the region of interest, where the region of interest
represents the injury according to the present invention; b)
determining the mask of the injury; c) determining a sub-tree below
the detected injury sub-image using the injury sub-image as the
root of the sub-tree; d) determining the soft threshold image
values for separating the core from the penumbra; e) comparing the
soft threshold image values inside the sub-tree to find the
penumbra and a mask of the penumbra; and f) determining the mask of
the core by subtracting the mask of the penumbra from the mask of
the injury. In one embodiment, the method further comprises
determining different gradations of the core and the penumbra.
[0010] In one embodiment, the method further comprises quantifying
the spatiotemporal evolution of an injury in living human
tissue.
[0011] In another embodiment, there is provided a method of
detecting the effects of endogenous or implanted stem cells on
living human tissue. The method comprises a) determining magnetic
resonance image values of labeled and implanted stem cells; b)
detecting the stem cells outside of the region of interest using a
method according to the present invention; and c) detecting the
stem cells inside of the region of interest using a method
according to the present invention. In one embodiment, the method
further comprises quantifying spatiotemporal activities of
implanted labeled stem cells in the living human tissue.
[0012] According to another embodiment of the present invention,
there is provided a method of detecting a core of an injury and
detecting a penumbra of an injury in living human tissue, and
distinguishing the core from the penumbra. The method comprises a)
configuring at least one processor to perform the functions of: 1)
detecting one sub(sub) image (the injury sub-image) in the
terminated Hierarchical Region Splitting Tree comprising the region
of interest, where the region of interest represents the injury
according to the present invention; 2) determining the mask of the
injury; 3) determining a sub-tree below the detected injury
sub-image using the injury sub-image as the root of the sub-tree;
4) determining the soft threshold image values for separating the
core from the penumbra; 5) comparing the soft threshold image
values inside the sub-tree to find the penumbra and a mask of the
penumbra; and 6) determining the mask of the core by subtracting
the mask of the penumbra from the mask of the injury. In one
embodiment, the method further comprises determining different
gradations of the core and the penumbra. In another embodiment, the
method further comprises quantifying the spatiotemporal evolution
of an injury in living human tissue.
[0013] According to one embodiment of the present invention, there
is provided a method of detecting the effects of endogenous or
implanted stem cells on living human tissue. The method comprises
a) configuring at least one processor to perform the functions of:
1) determining magnetic resonance image values of labeled and
implanted stem cells; 2) detecting the stem cells outside of the
region of interest using a method according the present invention;
and 3) detecting the stem cells inside of the region of interest
using a method according to the present invention. In one
embodiment, the method further comprises quantifying spatiotemporal
activities of implanted labeled stem cells in the living human
tissue.
[0014] According to another embodiment of the present invention,
there is provided a system for analyzing a medical image, where the
medical image comprises one or more than one region of interest.
The system comprises a) one or more than one processor; b) a
machine readable storage connected to the one or more than one
processor; c) a medical image comprising a set of actual image
values stored in the storage; d) a set of machine readable
instructions stored in the machine readable storage and operable on
the medical image; e) a user interface operably connected to the
set of computer instructions for transmitting one or more than one
command to the one or more than one processor; f) instructions
operably connected to the user interface for rescaling the actual
image values to produce corresponding rescaled image values and to
produce a rescaled image from the rescaled image values; g)
instructions operably connected to the user interface for deriving
a histogram of the rescaled image values; h) instructions operably
connected to the user interface for using the histogram to derive
an adaptive segmentation threshold that can be used to split the
rescaled image into two sub-images, a first sub-image with
intensities at or below the adaptive segmentation threshold and a
second sub-image with intensities above the adaptive segmentation
threshold, or a first sub-image with intensities below the adaptive
segmentation threshold and a second sub-image with intensities at
or above the adaptive segmentation threshold; i) instructions
operably connected to the user interface for using the adaptive
segmentation threshold to recursively split the rescaled image to
generate a Hierarchical Region Splitting Tree of sub(sub) images
based on consistency of the rescaled image values of the rescaled
image; j) instructions operably connected to the user interface for
terminating the recursive splitting of the sub(sub) images using
one or more than one predetermined criteria thereby completing the
Hierarchical Region Splitting Tree; k) instructions operably
connected to the user interface for identifying one sub(sub) image
in the Hierarchical Region Splitting Tree comprising the region of
interest; and l) a storage operably connected to the one or more
than one processor and the user interface for storing the resultant
Hierarchical Region Splitting Tree images.
DRAWINGS
[0015] These and other features, aspects and advantages of the
present invention will become better understood with regard to the
following description, appended claims, and accompanying drawings
where:
[0016] FIGS. 1A and 1B are block diagrams showing some steps in a
method according to the present invention of analyzing a medical
image according to the present invention;
[0017] FIG. 2 is a histogram of a rescaled apparent diffusion
coefficient image plotting rescaled apparent diffusion coefficient
(ADC) image values in the rescaled apparent diffusion coefficient
image on the x axis (in this case apparent diffusion coefficient
image values) versus frequency of each rescaled image value in the
rescaled image on the y axis;
[0018] FIG. 3 is a corresponding table of rescaled apparent
diffusion coefficient image values for the rescaled image showing
the adaptive segmentation threshold (Th) found for the histogram in
FIG. 2;
[0019] FIG. 4 is a schematic showing nomenclature rules for the
recursive splitting of a rescaled image;
[0020] FIG. 5 is a diagram of the top levels of a Hierarchical
Region Splitting Tree generated by the method of the present
invention for a magnetic resonance image which was generated from
the histogram of FIG. 2, where the level of image splitting is
shown on the right side of the diagram beginning with the first
image, Image 0, and the number of each sub-image and the range of
rescaled image values (v) present in the sub-image are shown below
each sub-image;
[0021] FIG. 6 is a block diagram showing some steps in another
method according to the present invention of detecting an
abnormality in living human tissue using a magnetic resonance image
as an example;
[0022] FIG. 7 and FIG. 8 show a histogram (left) of a rescaled
image (plotting apparent diffusion coefficient rescaled image
values in the rescaled image on the x axis versus frequency of each
rescaled image value in the rescaled image on the y axis) and the
part of the corresponding diagram (right) (Level 0, Level 1 and
Level 3) of the levels of a Hierarchical Region Splitting Tree
generated by the method of the present invention for a magnetic
resonance image which was generated from the histogram (left),
where the level of image splitting is shown on the right side of
the diagram beginning with the first image, Image 0, and the range
of rescaled image values present in the sub-image are shown below
each sub-image, where FIG. 6 is generated from the brain of a human
neonatal patient with a mild ischemic injury, and FIG. 7 is
generated from the brain of a human neonatal patient with a severe
ischemic injury;
[0023] FIG. 9 and FIG. 10 show a histogram (left) of a rescaled
image (plotting T2 relaxation time rescaled image values in the
rescaled image on the x axis versus frequency of each rescaled
image value in the rescaled image on the y axis) and the part of
the corresponding diagram (right) (Level 0, Level 1 and Level 2) of
the levels of a Hierarchical Region Splitting Tree generated by the
method of the present invention for a magnetic resonance image
which was generated from the histogram (left), where the level of
image splitting is shown on the right side of the diagram beginning
with the first image, Image 0, and the range of rescaled image
values present in the sub-image are shown below each sub-image,
where FIG. 9 is generated from a rat brain with a mild ischemic
injury, and FIG. 10 is generated from a rat brain with a severe
ischemic injury;
[0024] FIG. 11 is a diagram comparing volumetric results of
magnetic resonance images of injured brains using methods according
to the present invention (HRS) and using standard manual methods at
different injury-severities;
[0025] FIG. 12 is a diagram of core-penumbra injury detected
according to the present invention in an animal brain;
[0026] FIG. 13 is a diagram of core-penumbra injury detected
according to the present invention in a human neonatal brain;
and
[0027] FIG. 14 is a diagram depicting the detection of iron-labeled
stem cells in an ischemic animal brain over four weeks, where red
is the ischemic lesion, yellow is the iron-labeled murine neuronal
stem cells.
DESCRIPTION
[0028] According to one embodiment of the present invention, there
is provided a method of analyzing a medical image comprising a
region of interest, such as for example a medical image selected
from the group consisting of a computed tomography scan (CT scan),
a magnetic resonance image (MRI), a positron emission tomography
scan (PET scan) and an X-ray. In one embodiment, the region of
interest is a representation of an abnormality of living human
tissue, such as for example an injury to living human tissue, and
the method detects the representation of the abnormality. In a
preferred embodiment, the method additionally qualifies the region
of interest. In another embodiment, there is provided a method of
detecting a core of an injury and detecting a penumbra of an
abnormality of living human tissue, such as for example an injury
to living human tissue, and distinguishing the core from the
penumbra. According to one embodiment of the present invention,
there is provided a method of quantifying the spatiotemporal
evolution of an abnormality of living human tissue, such as for
example an injury to living human tissue. According to another
embodiment of the present invention, there is provided a method of
detecting the effects of endogenous or implanted neuronal stem
cells (NSCs) on living human tissue. In a preferred embodiment, the
living human tissue is a human brain.
[0029] The present method is an automated computational method
referred to as "Hierarchical Region Splitting" (HRS). Using a
magnetic resonance image as an example of a medical image,
Hierarchical Region Splitting advantageously analyzes a magnetic
resonance image approximately 100 times faster than analyzing the
magnetic resonance image by visual inspection and analysis of the
image by a trained technician which are the gold standard of
analyzing magnetic resonance images. Also advantageously,
Hierarchical Region Splitting can analyze medical images such as
magnetic resonance images and computed tomography scan in both two
dimensions and in three dimensions. Further advantageously,
Hierarchical Region Splitting does not require an atlas of normal
or diseased tissue for comparison, or depend on a probabilistic
disease model which are required by some other methods of analyzing
medical images such as magnetic resonance images. Hierarchical
Region Splitting can be used for a variety of analyses, including
detecting and quantifying an abnormality, such as for example an
ischemic injury, to a human brain or other tissue, qualifying the
abnormality, analyzing the internal characteristics of the
abnormality, quantifying the spatiotemporal evolution of the
abnormality, as well as determining the effects of endogenous or
implanted neuronal stem cells (NSCs) on living human tissue.
[0030] As used in this disclosure, except where the context
requires otherwise, the term "comprise" and variations of the term,
such as "comprising," "comprises" and "comprised" are not intended
to exclude other additives, components, integers or steps. Thus,
throughout this disclosure, unless the context requires otherwise,
the words "comprise," "comprising" and the like, are to be
construed in an inclusive sense as opposed to an exclusive sense,
that is to say, in the sense of "including, but not limited
to."
[0031] As used in this disclosure, except where the context
requires otherwise, the method steps disclosed and shown are not
intended to be limiting nor are they intended to indicate that each
step is essential to the method or that each step must occur in the
order disclosed.
[0032] As using in this disclosure, "injury" includes both
traumatic injury (such as for example gun shot wound) and
non-traumatic injury (such as for example ischemic stroke) as will
be understood by those with skill in the art with reference to this
disclosure.
[0033] As used in this disclosure, a "mask" of a region (such as
"region A" in a sub-image) is a black-and-white image referred to
as a "binary image" (usually of the same size as the original
image), where a pixel in white means that pixel is inside the
region and a pixel in black means that pixel is outside the region.
This binary image, when superimposed on the original medical image
(Level 0 of the Hierarchical Region Splitting Tree disclosed
below), reveals the sub-image containing region A only.
[0034] As used in this disclosure, the term "machine readable
medium" includes, but is not limited to portable or fixed storage
devices, optical storage devices, wireless channels and various
other mediums capable of storing, containing or carrying either
data, one or more than one instruction or both data and one or more
than one instruction.
[0035] As used in this disclosure, the term "computing device"
includes, but is not limited to computers, cellular telephones,
hand-held computers and other devices that are capable of executing
programmed instructions that are contained in a storage including
machine readable medium.
[0036] In the following disclosure, specific details are given to
provide a thorough understanding of the embodiments. However, it
will be understood by one of ordinary skill in the art that the
embodiments can be practiced without these specific details.
Well-known circuits, structures and techniques are not necessarily
shown in detail in order not to obscure the embodiments. For
example, circuits can be shown in block diagrams in order not to
obscure the embodiments in unnecessary detail.
[0037] Also, some embodiments are disclosed as a process that is
depicted as a flowchart, a flow diagram, a structure diagram, or a
block diagram. Although a flowchart discloses the operations as a
sequential process, many of the operations can be performed in
parallel or concurrently. In addition, the order of the operations
can be rearranged. A process is terminated when its operations are
completed. A process can correspond to a method, a function, a
procedure, a subroutine, a subprogram. When a process corresponds
to a function, termination of the process corresponds to a return
of the function to the calling function or the main function.
[0038] Moreover, a storage can represent one or more devices for
storing data, including read-only memory (ROM), random access
memory (RAM), magnetic disk storage mediums, optical storage
mediums, flash memory devices and other machine readable mediums
for storing information.
[0039] Furthermore, embodiments can be implemented by hardware,
software, firmware, middleware, microcode, or a combination of the
proceeding. When implemented in software, firmware, middleware or
microcode, the program code or code segments to perform the
necessary tasks can be stored in a machine-readable medium such as
a storage medium or other storage(s). One or more than one
processor can perform the necessary tasks in series or in parallel.
A code segment can represent a procedure, a function, a subprogram,
a program, a routine, a subroutine, a module, a software package, a
class, or a combination of instructions, data structures, or
program statements. A code segment can be coupled to another code
segment or a hardware circuit by passing or by receiving
information, data, arguments, parameters, or memory contents.
Information, arguments, parameters and data can be passed,
forwarded, or transmitted through a suitable means, such as for
example memory sharing, message passing, token passing, network
transmission.
[0040] According to one embodiment of the present invention, there
is provided a method of analyzing a medical image comprising a
region of interest, such as for example an image selected from the
group consisting of a computed tomography scan (CT scan), a
magnetic resonance image (MRI), a positron emission tomography scan
(PET scan) and an X-ray. In one embodiment, the region of interest
is a representation of an injury to living human tissue and the
method detects the representation of the injury, however, the
region of interest can be a representation of any abnormality as
will be understood by those with skill in the art with reference to
this disclosure. In one embodiment, the region of interest
comprises a representation of an abnormality to living human tissue
and the method both detects the representation of the abnormality
and qualifies the representation of the abnormality. In one
embodiment, the human tissue is selected from the group consisting
of heart, intestines, joints, kidneys, liver, lungs and spleen,
though any suitable tissue can be used as will be understood by
those with skill in the art with reference to this disclosure. In a
preferred embodiment, the living human tissue is brain. The method
will now be disclosed in greater detail with reference to magnetic
resonance imaging of the living human brain as an example of the
method, though corresponding steps can be used with other types of
medical images and other types of human tissue, as will be
understood by those with skill in the art with reference to this
disclosure.
[0041] According to one embodiment of the present invention, there
is provided a method of analyzing a medical image. Referring now to
FIGS. 1A and 1B, there are shown block diagrams showing some steps
in a method according to the present invention of analyzing a
medical image according to the present invention. As can be seen,
in one embodiment, the method comprises first, providing a medical
image. The medical image can be produced specifically for the
present method or can be previously produced for another purpose
and then used in the present method. The medical image provided can
be in a hard copy form (such as for example a film or print form)
or can be in a digital form. If the medical image provided is in a
hard copy form, the method further comprises preparing a digital
form of the medical image. The digital form of the medical image
comprises a set of actual image values.
[0042] Next, the method comprises rescaling the actual image values
to produce corresponding rescaled image values and a rescaled image
from the rescaled image values. In a preferred embodiment, the
medical image is a magnetic resonance image and the actual image
values are rescaled to fit in [0,255] unsigned 8-bit integer range;
however, other ranges can be used as will be understood by those
with skill in the art with reference to this disclosure. The
rescaling can be accomplished by any suitable method, as will be
understood by those with skill in the art with reference to this
disclosure. In one embodiment, rescaling is accomplished using the
following formula:
( rscVal - min RSCVAL ) max RSCVal - min RscVal ) = ( actVal - min
ActVal ) ( max ActVal - min ActVal ) ##EQU00001##
where minActVal is "minimum actual image value," maxActVal is
"maximum actual image value," minRscVal is "minimum rescaled image
value," and maxRscVal is "maximum rescaled image value." As an
example of the rescaling step of the method, according to the
present invention, the actual image values of a magnetic resonance
image were rescaled to generate a "rescaled image" (rscImg) by:
[0043] a) finding the "scale-factors" in the magnetic resonance
image ("Img"), where the scale factors consist of a "maximum image
value" (maxVal) and a "minimum image value" (minVal); [0044] b)
converting each "actual image value" (actVal) in the magnetic
resonance image to a corresponding "rescaled image value" (rscVal),
such as for example by using the formula:
[0044] rscVal=(actVal-minVal)/(maxVal-minVal)*255 [0045] thereby
producing a group of rescaled image values; [0046] c) saving the
scale-factors for converting each rescaled image value back to the
corresponding actual image value; and [0047] d) generating the
rescaled image of the magnetic resonance image using the group of
rescaled image values.
[0048] Then, the method comprises deriving a histogram of the
rescaled image values. Deriving the histogram can be accomplished
by: [0049] a) determining the frequency that a particular rescaled
image value in the range [0,255] appears in the rescaled image; and
[0050] b) producing an array [rescaled image values, frequency] in
the form of a histogram ("H") providing the following:
[0050] H(i)i.epsilon.[1,N]; where N=255
[0051] Next, the method comprises computing an adaptive
segmentation threshold (also referred to as a "histogram
shape-based image threshold") that can be used to split the gray
level image of the rescaled image into two sub-images, a first
sub-image with intensities at or below the adaptive segmentation
threshold and a second sub-image with intensities above the
adaptive segmentation threshold, or alternately a first sub-image
with intensities below the adaptive segmentation threshold and a
second sub-image with intensities at or above the adaptive
segmentation threshold, or alternately a first sub-image with
intensities below the adaptive segmentation threshold and a second
sub-image with intensities above the adaptive segmentation
threshold. The adaptive segmentation threshold can be computed
using any standard technique, as will be understood by those with
skill in the art with reference to this disclosure. In a preferred
embodiment, the adaptive segmentation threshold is computed using
Otsu's method, as will be understood by those with skill in the art
with reference to this disclosure. In summary, the histogram H(i)
is fit to a bimodal distribution (a function with two peaks), where
each of the two peaks corresponds to a distinct cluster of image
values from a relatively uniform region of the rescaled image, and
where each of the two peaks is distinct and separated enough to be
considered two different clusters. A valley (or trough) between the
two peaks is found that can separate the two peaks. The rescaled
image value of this valley is used as the adaptive segmentation
threshold ("Th"). Referring now to FIG. 2 and FIG. 3, there are
shown, respectively, a histogram of a rescaled apparent diffusion
coefficient (ADC) image plotting rescaled apparent diffusion
coefficient image values in the rescaled apparent diffusion
coefficient image on the x axis (in this case apparent diffusion
coefficient image values) versus frequency of each rescaled image
value in the rescaled image on the y axis (FIG. 2); and the
corresponding table of rescaled apparent diffusion coefficient
image values for the rescaled image showing the adaptive
segmentation threshold (Th) found for the histogram (FIG. 3).
[0052] The adaptive segmentation threshold (FIG. 3, right column)
was determined as follows: [0053] a) the histogram (H(i)) was
normalized to get the probabilistic distribution function, (pdf,
p(i));
[0053] p ( i ) = H ( i ) / i = 1 N H ( i ) ##EQU00002## where , N =
255 ##EQU00002.2## [0054] b) the cumulative distribution function
cdf, (.OMEGA.(i)) was found that is cumulative sum of the pdf,
p(I);
[0054] .OMEGA. ( i ) = j = 1 i p ( j ) ##EQU00003## [0055] c) the
cumulative weighted means (.mu.(i)) of the pdf, p(i) was found;
[0055] .mu. ( i ) = j = 1 i p ( j ) * j ##EQU00004## [0056] d) the
final weighted mean (.mu.t) was found;
[0056] .mu..sub.t=.mu.(N) where, N=255 [0057] e) Otsu's measure
(.sigma.b2(i)) was computed using the formula:
[0057] .sigma. b 2 ( i ) = [ .mu. t * .OMEGA. ( i ) - .mu. ( i ) ]
2 [ .OMEGA. ( i ) * ( 1 - .OMEGA. ( i ) ) ] ##EQU00005## [0058] f)
the mode (the number most frequently occurring in a series of
numbers; idx) was found in the series .sigma.b2(i). If there were
more than one mode, the mean of modes were used:
[0058] idx=mean[modes(.sigma..sub.b.sup.2)] [0059] g) the
normalized threshold (the adaptive segmentation threshold (Th), was
found using the following formula:
[0059] Th = ( idx - 1 ) ( N - 1 ) ##EQU00006## where , N = 255
##EQU00006.2##
[0060] Then, the method comprises recursively splitting the
resealed image using recursive bimodal segmentation to generate a
hierarchical tree (a "Hierarchical Region Splitting Tree") of
sub(sub) images based on consistency of the rescaled image values.
Referring now to FIG. 4, there is shown a schematic showing
nomenclature rules for recursive splitting of a rescaled image,
respectively. In a preferred embodiment, the rescaled image is
recursively split for example by: [0061] a) using the adaptive
segmentation threshold to split the rescaled image into two
sub-images: Lo_Img=(rscImg<Th) and Hi_Img=(rscImg.gtoreq.Th);
[0062] b) splitting the Lo_Img into 2 sub-sub-images using the
steps above (i. deriving a histogram, ii. computing an adaptive
segmentation threshold, and iii. splitting the rescaled image using
the adaptive segmentation threshold); [0063] c) splitting the
Hi_Img into 2 sub-sub-images using the steps above (i. deriving a
histogram, ii. computing an adaptive segmentation threshold, and
iii. splitting the rescaled image using the adaptive segmentation
threshold); [0064] d) continuing to recursively split the sub-sub
images generated by steps b) and c) to generate a hierarchical
structure of split image regions using corresponding steps, and
placing the images obtained by recursive splitting in a
Hierarchical Region Splitting Tree.
[0065] The sub-images are named following the rules below: [0066]
Rule 1: Level 0: Image 0 is the entire rescaled image forming the
root of the Hierarchical Region Splitting Tree. [0067] Rule 2:
Level 1, sub-images: Image 1 is Lo_Img, one of the first two
recursively split images from image 0 (placed as the left-leaf
(left-child) in the Hierarchical Region Splitting Tree). Image 2 is
Hi_img, the other of the first two recursively split images from
image 0 (placed as the right-leaf (right-child) in the Hierarchical
Region Splitting Tree). [0068] Rule 3: Level 2, sub-sub images:
Image 11 and Image 12 are the two recursively split images from
Image 1 (placed as the left-leaf (left-child) and the right-leaf
(right-child) of Image 1, respectively, in the Hierarchical Region
Splitting Tree) and a right-child "12"; Image 21 and Image 22 are
the two recursively split images from Image 2 (placed as the
left-leaf (left-child) and the right-leaf (right-child) of Image 2,
respectively, in the Hierarchical Region Splitting Tree); [0069]
Rule 4: The Lo_Img=(rscImg<Th) for each subsequent recursively
split parent image is named Image [number(s) of the parent followed
by a "1"]; and the Hi_Img=(rscImg.gtoreq.Th) for each subsequent
recursively split parent image is named Image [number(s) of the
parent followed by a "2"].
[0070] The further generated images continue to be split and named
according to Rule 4, as shown in FIG. 4, until predetermined
split-termination criteria (disclosed below) are fulfilled.
[0071] Next, the method comprises terminating the recursive
splitting of the sub(sub) images using one or more than one
predetermined criterion, and then identifying, preferably, one
sub(sub) image in the terminated Hierarchical Region Splitting Tree
which comprises the region of interest. In one embodiment, for
example, the medical image provided is a magnetic resonance image
of a human brain. Different types of brain tissues are differently
contrasted by magnetic resonance imaging. Therefore, brain regions
with uniform rescaled image values are expected to be from a single
type of brain tissue. As an example, in a preferred embodiment, the
magnetic resonance image provided is from human brain, and the
following predetermined criteria, Criteria 1 or Criteria 2 are used
to terminate the recursive splitting of the sub(sub) images to keep
the brain tissue regions of the images functionally meaningful and
the size of the Hierarchical Region Splitting Tree small: [0072]
Criteria 1: area threshold=50 pixels or approximately 2 ml in
volume, (AreaTh=50 pixels or approximately 2 ml in volume); when
the individual connected regions for a split image are small
(area<50 pixels or approximately 2 ml volume), the rescaled
image values are probably from a single brain region; or [0073]
Criteria 2: standard deviation threshold=10 rscVals (StdDevTh=10
rscVals) and kurtosis threshold=1.5 (KurtTh=1.5); when both the
rescaled image values have a low standard deviation (STD<10
rscVals) and the corresponding histogram for a split image has a
single very sharp peak (kurtosis<1.5); where the kurtosis image
value of a Gaussian Normal distribution is 3, the rescaled image
values are probably from a single brain region.
[0074] Referring now to FIG. 5, there is shown a diagram of the top
levels of a Hierarchical Region Splitting Tree generated by the
method of the present invention for a magnetic resonance image
which was generated from the histogram of FIG. 2, where the level
of image splitting is shown on the right side of the diagram
beginning with the first image, Image 0, and the number of each
sub-image and the range of rescaled image values (v) present in the
sub-image are shown below each sub-image.
[0075] In one embodiment, the method further comprises performing a
secondary rescaling (a scaling back) of some or all of the rescaled
sub-images in the Hierarchical Region Splitting Tree generated from
the medical image provided back to the actual image values present
in the medical image provided to create secondary rescaled medical
images. This secondary rescaling of sub-images generates
scaled-back sub-images having the same range of image values as the
range of image values present in the medical image provided and can
be used to generate a Hierarchical Region Splitting Tree using the
actual image values. In one embodiment, the secondary rescaling of
each sub-image is performed using a formula that is the counterpart
of the formula in the rescaling step, as follows:
actVal=[(rscVal/255)*(maxVal-minVal)]+minVal
[0076] According to another embodiment of the present invention,
there is provided a method of detecting an abnormality in living
human tissue. The method comprises analyzing a medical image
according to the present invention, where the medical image
comprises a representation of the abnormality in the living human
tissue. In one embodiment, the method further comprises quantifying
the abnormality in the living human tissue.
[0077] When imaged on a magnetic resonance image, injured brain
tissues tend to have signal image values in particular magnetic
resonance image modalities that are distinct from non-injured brain
tissues, and therefore injured brain tissues tend to have good
contrast with the adjacent non-injured brain tissues in magnetic
resonance images, which makes detection and quantification of
injured brain tissues from magnetic resonance images using the
present method particularly effective. For example, referring again
to FIG. 5, using the method of analyzing a medical image as
disclosed in this disclosure, a magnetic resonance image of a
living human brain was analyzed and the injured brain region that
suffered an ischemic injury was detected as shown in Image 11 in
level 2 as a hypo-intense ischemic injury region that is clearly
visible in Image 11.
[0078] Referring now to FIG. 6, there is shown a block diagram
showing some steps in a method according to the present invention
of detecting an abnormality in living human tissue using a magnetic
resonance image as an example. As can be seen, the method of
detecting an abnormality in living human tissue comprises, first,
analyzing a medical image as disclosed in this disclosure, where
the medical image comprises a representation of the abnormality in
the living human tissue. The method of detecting an abnormality in
living human tissue will now be disclosed in further detail using a
magnetic resonance image as an example of the medical image, and an
injury to brain tissue as an example of an abnormality in the
living human tissue; however, as will be understood by those with
skill in the art with reference to this disclosure, the medical
image can be any suitable medical image, such as for example a
computed tomography scan (CT scan), a magnetic resonance image
(MRI), a positron emission tomography scan (PET scan) and an X-ray,
and the abnormality can be any suitable abnormality, such as for
example a genetic malformation and an injury.
[0079] Next, the method comprises determining an image value
(MeanTh) or a set of image values (MeanThs) of actual image values
in a medical image (after the secondary rescaling back to the Img
image values), where the MeanTh or MeanThs determined identifies
the abnormality represented in the medical image, such as a region
of brain tissue as injured brain tissue, for the modality being
used to generate the medical image provided, where the region of
injured brain tissue has an actual image value that is less than
the MeanTh, greater than the MeanTh, or within the set of MeanThs,
and where the MeanTh(s) comprises a type(s) and an amount(s). The
MeanTh(s) are sometimes called a soft threshold image value(s).
[0080] In one embodiment, determination of the MeanTh(s) is made
before the step of providing the medical image. In another
embodiment, determination of the MeanTh(s) is made after the step
of providing the medical image. In one embodiment, determining the
MeanTh(s) comprises consulting references containing these
MeanTh(s), as will be understood by those with skill in the art
with reference to this disclosure. In another embodiment,
determining the MeanTh(s) comprises performing research to
ascertain the MeanTh(s), as will be understood by those with skill
in the art with reference to this disclosure. In one embodiment,
the medical image is a magnetic resonance image and the modality
being used to generate the medical image provided is selected from
the group consisting of an apparent diffusion coefficient (ADC)
map, and a magnetic susceptibility map and a T2 map, though any
suitable modality can be used as will be understood by those with
skill in the art with reference to this disclosure. In one
embodiment, the type of the MeanTh(s) determined is selected from
the group consisting of diffusion coefficient, magnetic
susceptibility and T2 relaxation time, though any suitable type of
the MeanTh(s) can be used as will be understood by those with skill
in the art with reference to this disclosure.
[0081] Then, the method comprises selecting a "relational operator"
selected from the group consisting of "less than" (the MeanTh),
"greater than" (the MeanTh), and "within" (the MeanThs), where the
relation operator determined indicates the relationship of the
MeanTh(s) determined to the mean of the actual image values in the
medical image (the secondarily rescaled image values, that is, the
rescaled image values after scaling back to actual image values) of
the injured brain tissue for the modality being used to generate
the medical image (in this example, a magnetic resonance image)
provided. This relationship is used to determine whether the
tissues imaged in a sub-image of the medical image constitutes
injured brain tissue for the medical image modality being used.
[0082] Next, the method comprises comparing the MeanTh(s) to the
average image value of each of the scaled-back sub-images of the
Hierarchical Region Splitting Tree sequentially starting from the
top level (level 0) of the Hierarchical Region Splitting Tree and
then downwards through level 1, level 2, and subsequent levels
until reaching the first scaled-back sub-image "A" in the
Hierarchical Region Splitting Tree that has an average actual
magnetic resonance image value that satisfies the relational
operator with the MeanTh(s), where the sub-image "A" comprises the
abnormality (such as an ischemic injury) in the brain.
[0083] For example, from published works, in the modality of an
apparent diffusion coefficient (ADC) map for a magnetic resonance
image, ischemic injury to the brain has a MeanTh in rescaled
apparent diffusion coefficient image value of 80 and a relational
operator of "less than." Referring again to FIG. 5, as can be seen,
Image `11` has a mean image value (for apparent diffusion
coefficient) of 76, an image value that is "less than" the
determined MeanTh of 80 and, therefore, Image `11` is detected as
comprising an abnormality in the injured living human tissue (here
ischemic injured brain tissue). (As can be appreciated, the actual
adaptive threshold (95) used to create Image `11` from Image 1 and
an actual mean of this image sub-region (76) are different from the
MeanTh (80) used to detect the abnormality in the living human
tissue.)
[0084] For example, in ADC maps (magnetic resonance images) of
ischemically-injured brain, when the mean ADC image value of a
scaled-back ADC sub-image "A" is "less than" the MeanTh in ADC unit
(mm/sec.sup.2), that sub-image "A" is delineated as ischemic injury
based on the magnetic resonance image modality of diffusion
coefficients. Similarly, in T2 maps of is chemically-injured brain,
when the mean T2 relaxation time (millisecond) of a scaled-back T2
sub-image "B" is "greater than" the MeanTh (in milliseconds), then
sub-image "B" is delineated as ischemic injury based on the
magnetic resonance image modality of T2 relaxation time.
[0085] When the relational operator is "within," there are two
MeanThs, and an injury is determined to be present in the sub-image
where the mean actual image value of a scaled-back sub-image is
between the two MeanThs. Determination that a scaled-back sub-image
is between the two MeanThs (MeanTh1 and MeanTh2) can be made by
detecting the first complementary sub image as follows (where
MeanTh1=100 and MeanTh2=180 in this example): [0086] a) determining
the mask (mask1) of the region for "less than" MeanTh1; [0087] b)
determining the mask (mask2) of the region for "greater than"
MeanTh2; [0088] c) determining the union of the masks [mask12=AND
(mask1, mask2)] from step a) and [0089] d) this is the
complementary region of what is being searched; [0090] e)
determining the mask (mask0) of the entire brain region from Image
`0`; and [0091] f) determining the mask of the detected injury
region (injuryMask) by subtracting mask12 from mask0 (i.e.,
mask*=mask0-mask12).
[0092] In one embodiment, the method further comprises cleaning the
mask of the detected injury region (injuryMask) using morphological
opening, closing and cleaning to remove small outlier regions to
generate a cleaned injuryMask, as will be understood by those with
skill in the art with reference to this disclosure. In another
embodiment, the method further comprises using the cleaned
injuryMask for further morphological quantifications (such as for
example, area/volume, 2D/3D shape, boundary contour/surface,
anatomical location, eigenvectors/image values, major/minor axes,
orientation, compactness), as will be understood by those with
skill in the art with respect to this disclosure.
[0093] Referring now to FIG. 7 and FIG. 8, each Figure shows a
histogram (left) of a rescaled image (plotting apparent diffusion
coefficient rescaled image values in the rescaled image on the x
axis versus frequency of each rescaled image value in the rescaled
image on the y axis) and the part of the corresponding diagram
(right) (Level 0, Level 1 and Level 3) of the levels of a
Hierarchical Region Splitting Tree generated by the method of the
present invention for a magnetic resonance image which was
generated from the histogram (left), where the level of image
splitting is shown on the right side of the diagram beginning with
the first image, Image 0, and the range of rescaled image values
present in the sub-image are shown below each sub-image, where FIG.
7 is generated from the brain of a human neonatal patient with a
mild ischemic injury, and FIG. 8 is generated from the brain of a
human neonatal patient with a severe ischemic injury. As can be
seen, the abnormality is detected in the left-most image of Level 3
in both diagrams because the left-most image of Level 3 is detected
as the first sub-image that has a mean scaled-back image value (the
scaled-back image value being the same as the original diffusion
coefficient image value) that satisfies the relational operator
("less than" in these examples) and the MeanTh (0.16 mm2/sec
diffusion for ischemic injury) previously determined.
[0094] Referring now to FIG. 9 and FIG. 10, each Figure shows a
histogram (left) of a rescaled image (plotting rescaled image
values for T2 relaxation time in the rescaled image on the x axis
versus frequency of each rescaled image value in the rescaled image
on the y axis) and the part of the corresponding diagram (right)
(Level 0, Level 1 and Level 2) of the levels of a Hierarchical
Region Splitting Tree generated by the method of the present
invention for a magnetic resonance image which was generated from
the histogram (left), where the level of image splitting is shown
on the right side of the diagram beginning with the first image,
Image 0, and the range of rescaled image values present in the
sub-image are shown below each sub-image, where FIG. 9 is generated
from a rat brain with a mild ischemic injury, and FIG. 10 is
generated from a rat brain with a severe ischemic injury. As can be
seen, the abnormality is detected in the right image of Level 1 in
both diagrams because the right image of Level 1 is detected as the
first sub-image that has a mean scaled-back image value (the
scaled-back image value being the same as the original T2
relaxation time image value) that satisfies the relational operator
("greater than" in these examples) and the MeanTh (180 millisecond
T2 relaxation time for ischemic injury) previously determined.
[0095] Referring now to FIG. 11, there is shown a diagram comparing
volumetric results of magnetic resonance images of injured brains
using methods according to the present invention (HRS) and using
standard manual methods at different injury-severities. T2WI from
mild (<15% lesion), moderate (15-35%) and severe (>35%)
injuries are shown at the level of the horizontal line. The
percentage of the lesion volume compared to the entire brain is
shown below the 3D volumes. As can be seen, manually detected
lesions in 2D (T2WI, row 1) and in 3D (row 2) and lesions detected
according to the present invention (HRS) in 2D (T2WI, row 3) and in
3D (row 4) were similar between both methods, demonstrating that
results using the present method (HRS) correlated accurately with
results using the standard manual method.
[0096] According to another embodiment of the present invention,
there is provided a method of detecting a core of an injury and
detecting a penumbra of an injury in living human tissue, and
distinguishing the core from the penumbra. The method comprises
analyzing the medical image according to the present invention. In
one embodiment, the human tissue is selected from the group
consisting of heart, intestines, joints, kidneys, liver, lungs and
spleen, though any suitable tissue can be used as will be
understood by those with skill in the art with reference to this
disclosure. In a preferred embodiment, the living human tissue is
brain. The method will now be disclosed in greater detail with
reference to magnetic resonance imaging of the living human brain
as an example of the method, though corresponding steps can be used
with other types of medical images and other types of human tissue,
as will be understood by those with skill in the art with reference
to this disclosure.
[0097] The "core" of an injury is the area or volume that contains
tissues that are dead and completely irrecoverable. The "penumbra"
of an injury is the area or volume that contains tissue that is not
dead but that is affected by the injury, where some of the tissue
is recoverable. The penumbra is generally located adjacent to or
around the core. Outside of the core and penumbra are normal
healthy tissues. The water content (such as for example as
determined by T2 maps) and water mobility (such as for example as
determined by apparent diffusion coefficient maps) in magnetic
resonance images are generally different for the core as compared
to the penumbra, and the water content and water mobility in
magnetic resonance images are generally different for both the core
and the penumbra as compared to normal healthy tissues.
[0098] In one embodiment, the method according to the present
invention of detecting a core of an injury and detecting a penumbra
of an injury in living human tissue, and distinguishing the core
from the penumbra is accomplished by: [0099] a) detecting one
sub(sub) image (designated the "injury sub-image") comprising the
region of interest in the terminated Hierarchical Region Splitting
Tree, where the region of interest represents the injury according
to the present invention; [0100] b) determining the mask of the
injury ("InjuryMask"); [0101] c) determining a sub-tree below the
detected injury sub-image using the injury sub-image as the root of
the sub-tree; [0102] d) determining the MeanThs (soft threshold
image values), here referred to as the "MeanThPnmb," which is the
image value for separating the core from the penumbra by
determining the water content (such as for example as determined by
T2 maps) and water mobility (such as for example as determined by
apparent diffusion coefficient maps) in magnetic resonance images
associated with the core and the penumbra of a particular type of
injury from published sources or from expert knowledge; [0103] e)
comparing the MeanThPnmb inside the sub-tree to find the penumbra
and the mask of the penumbra ("PnmbMask"); and [0104] f)
determining the mask of the core ("CoreMask") by subtracting the
PnmbMask from InjuryMask (that is, the CoreMask=the InjuryMask-the
PnmbMask). (Shown in red regions in FIG. 12 and FIG. 13 where the
Figures are in color and in the lighter gray where the Figures are
not in color).
[0105] For example, referring again to FIG. 5, Image 11 is the
injury sub-image and the sub-tree is Image 111 and Image 112 (as
shown in FIG. 5), and sub-images below Image 111 and Image 112
(that are not shown in FIG. 5). Known T2 relaxation time image
values for neonatal ischemic injury are greater than 200 ms for the
core, and 160 ms<T2 relaxation time<200 ms for the penumbra,
and known apparent diffusion coefficient is less than
0.25.times.10.sup.-3 mm.sup.2/sec for the core and
0.25.times.10.sup.-3 mm2/sec<apparent diffusion
coefficient<0.50.times.10.sup.-3 mm.sup.2/sec for the penumbra.
Hence, using the image values in step c) and the mean T2 relaxation
time ("meanT2") and/or mean apparent diffusion coefficient
("meanADC") of a sub-image in the sub-tree from step b), the
penumbra is decided by meanT2<MeanThPnmb=200 ms, and by apparent
diffusion coefficient meanADC>MeanThPnmb=0.25.times.10.sup.-3
mm.sup.2/sec.
[0106] In one embodiment, the method further comprises determining
different gradations (quantitative measures) of the core and the
penumbra using steps corresponding to other embodiments of the
method disclosed in this disclosure as will be understood by those
with skill in the art with reference to this disclosure. The
different quantitative measures of the core and the penumbra are
useful for better pathological temporal monitoring and therapeutic
intervention, even if there is no scientific term yet associated
with the quantitative measures (such as core and penumbra). In
summary, the corresponding steps for determining the different
gradations of the core and penumbra are as follows: [0107] a)
providing the sub-tree of the HRS tree; [0108] b) identifying small
ranges in-between the entire signal range of the injury sub-image
in the root image that correspond to corresponding gradations of
the core and penumbra; and [0109] c) computing the different
gradations.
[0110] In one embodiment, the method further comprises clustering
multiple sub-images with the same range (from different branches
and different levels of the HRS tree) to get unified sub-region
structures of the injury.
[0111] Referring now to FIG. 12, there is shown a diagram of
core-penumbra injury detected according to the present invention in
an animal brain. As can be seen, using the present method, core was
detected as red areas in row 2, and penumbra was detected as blue
areas in row 2 using T2 maps (row 1). Further, using the present
method, even finer gradations of the injury beyond the simple
core-penumbra separation were identified in row 3, where the red:
T2>220; magenta: 200<T2<220; yellow: 190<T2<200;
blue: 180<T2<190; cyan: 170<T2<180; green:
150<T2<170; and white: 140<T2<150 visually depict the
finer gradations, where the Figures are in color and in varying
shades of gray where the Figures are not in color.
[0112] Referring now to FIG. 13, there is shown a diagram of
core-penumbra injury detected according to the present invention in
a human neonatal brain. As can be seen, using the present method,
core was detected as red areas and penumbra as blue areas in row 2
and row 5 using apparent diffusion coefficient maps (row 1 and row
4). Further, using the present method, even finer gradations of the
injury beyond meanADC is mean apparent diffusion coefficient of a
sub-image in the sub-tree from step a) above and pseudo-colors used
to visualize the finer gradation are as follows: [0113] a) red:
meanADC<0.10.times.10-3 mm.sup.2/sec; [0114] b) magenta:
0.10.times.10.sup.-3
mm.sup.2/sec<meanADC<0.20.times.10.sup.-3 mm.sup.2/sec;
[0115] c) yellow: 0.20.times.10.sup.-3
mm.sup.2/sec<meanADC<0.25.times.10.sup.-3 mm.sup.2/sec;
[0116] d) blue: 0.25.times.10.sup.-3
mm.sup.2/sec<meanADC<0.30.times.10.sup.-3 mm.sup.2/sec;
[0117] e) cyan: 0.30.times.10.sup.-3
mm.sup.2/sec<meanADC<0.35.times.10.sup.-3 mm.sup.2/sec;
[0118] f) green: 0.35.times.10.sup.-3
mm.sup.2/sec<meanADC<0.40.times.10.sup.-3 mm.sup.2/sec; and
[0119] g) white: 0.40.times.10.sup.-3
mm.sup.2/sec<meanADC<0.50.times.10.sup.-3 mm.sup.2/sec. where
the Figures are in color and in varying shades of gray where the
Figures are not in color. The same techniques and range of image
values are used for FIGS. 12 and 13, where only finer gradations of
the injury are shown for an injured animal brain and injured human
brain, respectively.
[0120] As will be understood by those with skill in the art with
reference to this disclosure, injuries usually evolve spatially and
temporally as does many other types of abnormalities such as
genetic abnormality, that is, the anatomical location and extent of
an injury changes over time after the injury and usually involve an
initial degeneration process followed by a recovery process.
According to one embodiment of the present invention, there is
provided a method of quantifying the spatiotemporal evolution of an
injury in living human tissue. In one embodiment, the method
comprises: [0121] a) using an established (preferably age-matched)
atlas of the injured tissue, such as for example an atlas of
injured brain, and a standard (manual or automatic) co-registration
method to overlap the two-dimensional or three-dimensional magnetic
resonance image onto the two-dimensional or three-dimensional atlas
of the injured tissue; [0122] b) determining the anatomical regions
involved in the injury at different granular levels (such as for
example the entire injured region, the core and penumbra, or finer
gradations of the injured region); [0123] c) quantifying different
features specific to the spatial overlaps; and [0124] d) using
longitudinal imaging data (such as for example neuro-imaging data
in the case of brain injury) to reveal temporal variations of the
spatial features in the quantifying step.
[0125] By way of example only, such quantifying of the
spatiotemporal evolution of an injury comprises quantifying the
mean T2 image value of the injured tissues over time, how the
volume of an injury changes over time, and how an injured tissue
recovers over time.
[0126] According to another embodiment of the present invention,
there is provided a method of detecting the effects of endogenous
or implanted stem cells (such as for example neuronal stem cells
(NSCs)) on living human tissue. In one embodiment, the method
involves the automated detection and quantitative monitoring of
stem cells. Stem cells that have been labeled with iron are visible
on various medical imaging techniques, such as for example on
magnetic resonance images in the modality of T2 maps and
susceptibility weighted imaging (SWI) maps where the iron labeled
stem cells appear as dark (hypo-intense) small clusters on the
magnetic resonance image. Markers other than iron that allow stem
cells to be distinguished from surrounding tissues are also useful
for this method, as will be understood by those with skill in the
art with reference to this disclosure. In the present invention,
the methods disclosed in this disclosure are used to monitor
labeled stem cells. In one embodiment, the method comprises: [0127]
a) determining the magnetic resonance image values of the labeled
and implanted stem cells, such as for example using the published
work or original research. For example, in T2 maps, iron labeled
neuronal stem cells outside of an injury usually have pixels/voxels
with a T2 relaxation time of less than 50 ms (that is, the soft
approximate threshold in T2 maps for iron labeled neuronal stem
cells is MeanThNSCout=50 ms for iron labeled neuronal stem cells
detection). Alternately, iron labeled neuronal stem cells inside an
injury can be different from iron labeled neuronal stem cells
outside of the injury due to superimposition of injury-contrast and
iron labeled neuronal stem cells-contrast and voxel-averaging
effect in the magnetic resonance image, where ischemic injuries are
bright (hyper-intense) and iron-labeled NSCs are dark
(hypo-intense) in T2 maps. Hence, the corresponding approximate
threshold the "MeanTh NSCin" is determined by the equation:
[0127] meanThNSCin=meanThNSCout/meanNABM*meanlnjury
where "meanNABM" and "meanlnjury" are the actual mean T2 image
values of the normal area brain matter (NABM) and the injury,
respectively. This assumes that the contrast ratio between
iron-labeled neuronal stem cells and the surrounding tissues are
same in T2 maps, whether the neuronal stem cells are outside of or
inside the injury. [0128] b) Neuronal Stem Cells Detection Outside
Injury: From the Hierarchical Region Splitting Tree generated
according to the present invention, the sub-regions with a mean MRI
image value less than MeanThNSC1 are found, and the mask of the
detected stem cells (nscMask) is found, again according to the
present invention. Referring now to FIG. 14, there is shown a
diagram depicting the detection of iron-labeled stem cells in an
ischemic animal brain over four weeks, where red is the ischemic
lesion, yellow is the iron-labeled murine neuronal stem cells,
where the Figures are in color and in varying shades of gray where
the Figures are not in color. As can be seen, the iron labeled
neuronal stem cells were detected in the left-most strand
(containing sub-images `0`-`1`-`11`-`111`-`1111`- and so on) of the
T2-map-based Hierarchical Region Splitting Tree near the bottom of
the entire tree (the entire HRS tree is not shown). [0129] c)
Neuronal Stem Cells Detection Inside Injury: In this case, HRS
sub-tree is considered below the detected injury in the magnetic
resonance image, such as in T2 maps, and the sub-region of the
detected injury (injuryMask) is found that has mean magnetic
resonance image value less than the approximate threshold
"MeanThNSCin." This sub-region of the injury is identified as
neuronal stem cells inside the injury. In general, the neuronal
stem cell regions are found at the left-most strand of the sub-tree
with detected injury as the root. [0130] d) quantify different
features morphological quantifications of the "nscMask." In a
preferred embodiment, no morphological cleaning is done as neuronal
stem cells clusters are sometimes very small in size. In one
embodiment, the feature quantified is selected from the group
consisting of (actual) mean, anatomical location, area/volume,
2D/3D shape, compactness, standard deviation and weighted
centroid.
[0131] Stem cells are attracted by signals from an injury region
and the stem cells migrate, proliferate, differentiate and take
part in recovery from the injury, including injury to the brain.
Quantification of the stem cells' activities (in vivo) is currently
performed by time-consuming and subjective visual/manual methods.
According to another embodiment of the present invention, there is
provided a method of quantifying spatiotemporal activities of
implanted labeled stem cells in living human tissue, including
human brain. In one embodiment, the method comprises steps
corresponding to steps from methods disclosed in this disclosure.
In summary, shape, proximity and area-similarity based matching is
done to track specific stem cells cluster over space and time.
Migrations of stem cells are computed by location changes of the
magnetic resonance image-signal-weighted centroid of the same stem
cells cluster over time. Direction, speed or both direction and
speed can be determined. Proliferations of stem cells are computed
by the expansion and compression of the area or volume of a
particular stem cells cluster, where directional preferences in
proliferation are computed by changing shapes over time. Higher
order statistics of the migration and proliferation (such as for
example rate of change of migration and rate of change of
proliferation) are also computed for detailed stem cells
activities. Final locations of the stem cells are computed by
determining the "leading edge," that is, the farthest pixel/voxel
of the stem cells cluster from the implantation site. As different
stem cells clusters take different paths towards the injury site,
path-specific stem cells activities are quantified and compared for
to allow monitoring of stem cell therapy.
[0132] According to another embodiment of the present invention,
there is provided a method of quantifying the interaction between
injury evolution and stem-cell activities in living human tissue,
including human brain. In one embodiment, the method comprises
steps corresponding to steps from methods disclosed in this
disclosure.
[0133] Although the present invention has been discussed in
considerable detail with reference to certain preferred
embodiments, other embodiments are possible. Therefore, the scope
of the appended claims should not be limited to the description of
preferred embodiments contained in this disclosure. All references
cited herein are incorporated by reference in their entirety.
* * * * *