U.S. patent application number 09/782724 was filed with the patent office on 2002-08-15 for method for stitching partial radiation images to reconstruct a full image.
This patent application is currently assigned to Eastman Kodak company. Invention is credited to Foos, David H., Steklenski, David J., Wang, Xiaohui.
Application Number | 20020109113 09/782724 |
Document ID | / |
Family ID | 25126976 |
Filed Date | 2002-08-15 |
United States Patent
Application |
20020109113 |
Kind Code |
A1 |
Wang, Xiaohui ; et
al. |
August 15, 2002 |
METHOD FOR STITCHING PARTIAL RADIATION IMAGES TO RECONSTRUCT A FULL
IMAGE
Abstract
A method of forming a composite image from first and second
digital images formed by recording first and second contiguous
segments of a larger radiographic image in first and second
overlapping storage phosphor members, exposed to a source of X-rays
wherein the image content in the overlapped region is the same in
both images and the end edge of the first member is present both on
the first image and as a shadow edge in the second image, the
method comprising: correcting for geometric distortion in the first
and second digital images, determining any rotational displacement
and any vertical displacement between the first and second images
by matching the first member end edge in the first image to its
shadow in the second image; correcting for image orientation based
on any said rotational displacement; determining any horizontal
displacement between the first and second images by correlating the
image content in the overlapped region of the first and second
images; and stitching said first and second images together along
the first member end edge based on any said horizontal and vertical
displacements.
Inventors: |
Wang, Xiaohui; (Pittsford,
NY) ; Foos, David H.; (Rochester, NY) ;
Steklenski, David J.; (Rochester, NY) |
Correspondence
Address: |
Thomas H. Close
Patent Legal Staff
Eastman Kodak Company
343 State Street
Rochester
NY
14650-2201
US
|
Assignee: |
Eastman Kodak company
|
Family ID: |
25126976 |
Appl. No.: |
09/782724 |
Filed: |
February 13, 2001 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
09782724 |
Feb 13, 2001 |
|
|
|
09742509 |
Dec 20, 2000 |
|
|
|
Current U.S.
Class: |
250/584 |
Current CPC
Class: |
G01T 1/2014 20130101;
G03B 42/04 20130101 |
Class at
Publication: |
250/584 |
International
Class: |
G03B 042/02 |
Claims
What is claimed is:
1. A method of forming a composite image from first and second
digital images formed by recording first and second contiguous
segments of a larger radiographic image in first and second
overlapping storage phosphor members, exposed to a source of X-rays
wherein the image content in the overlapped region is the same in
both images and the end edge of said first member is present both
on said first image and as a shadow edge in said second image, said
method comprising: correcting for geometric distortion in said
first and second digital images; determining any rotational
displacement and any vertical displacement between said first and
second images by matching said first member end edge in said first
image to its shadow in said second image; correcting for image
orientation based on any said rotational displacement; determining
any horizontal displacement between said first and second images by
correlating said image content in said overlapped region of said
first and second images; and stitching said first and second images
together along said first member end edge based on any said
horizontal and vertical displacements.
2. The method of claim 1 herein said first and second digital
images include a matrix of pixels and wherein said geometric
distortion correction of said images varies over the length of said
images.
3. The method of claim 2 wherein said geometric distortion
correction increases in said overlapped region of said image.
4. The method of claim 2 wherein said geometric distortion
correction is a function of the distance between said first and
second overlapping storage phosphor plates and said source of
X-rays.
5. The method of claim 1 wherein said first member end edge is
located within said first and second images by detecting pixel
value discontinuity in said overlap region of said images.
6. The method of claim 5 wherein said pixel value discontinuity
detection is carried out by (1) computing all the significant edge
transition pixels in the proximity of the first member end edge
location, and (2) performing line delineation of the end edge
pixels.
7. The method of claim 1 wherein said correcting for image
orientation is carried out by one of, (1) rotating said first
digital image while keeping said second digital image unchanged,
(2) rotating said second digital image while keeping said first
digital image unchanged, or (3) rotating both said first and second
digital images relative to one another.
8. The method of claim 1 wherein said determining any horizontal
displacement is carried out (1) by defining and extracting the
first image overlap region, (2) by defining and extracting the
second image overlap region, and (3) calculating any horizontal
displacement by image correlation.
Description
CROSS REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit under 35 USC .sctn.120
of the earlier filing date of U.S. patent application Ser. No.:
09/742,509, filed Dec. 20, 2000, entitled Elongated Computed
Radiography Cassette
FIELD OF THE INVENTION
[0002] This invention relates in general to digital radiography,
and in particular to the imaging of a long human body part, such as
the spine or legs, using a storage phosphor-based computed
radiography system.
BACKGROUND OF THE INVENTION
[0003] When a long segment of the human body is imaged using the
conventional screen-film technique, special cassettes and films of
extended length are used, such as 30.times.90 cm and 35.times.105
cm. As medical institutions are migrating from analog screen-film
systems to digital modalities, such as computed radiography (CR),
these types of exams impose a significant challenge. This is
because the size of digital detector is limited. For example, the
largest CR storage phosphor cassette from several major CR vendors
is limited to 35.times.43 cm, which can only image a partial of the
body part at a time. To address this problem, a method has been
proposed that staggers several storage phosphor cassettes together
in a specially made cassette holder (U.S. Pat. No. 5,986,279,
European Patent App. EP0866342A1, EP 0919856A1, and EP0919858A1).
All the cassettes are exposed in a single x-ray exposure. Then
image processing is applied to stitch all the partial images
together. The advantage is that the method is compatible with the
current CR readers. However, a pattern of reference markers needs
to be imaged simultaneously with the patient in order to achieve
precise geometric registration of the partial images. The shadow of
the reference markers may obscure diagnostically important
information in the stitched image. Also because of the overlapping
of the cassettes, the metallic cassette frames introduce wide
shadow artifacts in the resultant image that are sometimes
objectionable. Moreover, the cassette holder is quite heavy and is
typically mounted in a fixed position, which limits the users from
moving it up and down for exact patient positioning. The cassette
holder is bulky and does not conform to ISO/ANSI standards, which
means that it can not be placed in the bucky grid holder that is
designed for the current screen-film systems. U.S. patent
application Ser. No.: 09/742,509 filed Dec. 20, 2000, discloses a
method that is based on an extended length cassette with two
35.times.43 cm phosphor screens built inside. The two phosphor
screens are slightly overlapped in the center of the cassette
(FIGS. 1-3). The overall cassette size is about 35.times.85 cm,
which nearly doubles the current largest cassette size and allows a
fairly long segment of the human body to be imaged at a single
exposure. The information recorded in either phosphor screen bears
part of the desired final image.
[0004] During the readout process, one end of the cassette is
placed in the CR reader and the first phosphor screen is scanned
and stored, the cassette is then removed from the reader and
inverted to allow the second phosphor screen to be read in the same
manner as the first. The two images can then be processed into a
composite full image if so desired. The length of the cassette can
be designed to be shorter or longer in order to follow the ISO/ANSI
standard, such as 36" and 51" inch long. The maximum cassette
length is approximately twice the maximum allowable scan length of
the CR reader.
[0005] Special digital image processing is required to construct a
composite full image from the front and back images that are
obtained from the two individual phosphor screens. The two phosphor
screens are packed and partially overlapped inside the single
cassette and are therefore not coplanar. This causes the image of
the body part to be magnified differently for different locations
in the cassette, and a demagnification operation is required as
part of the process of registering the front and back images. In
addition, the two phosphor screens will not be perfectly aligned
inside the cassette, and there are translation and rotational
displacements introduced by the CR reader during the image readout
process. As a result, the placement of the pixels from the front
and back images will not be perfectly aligned, and the images will
require rotation and translation compensation. The aforementioned
image registration processing can be accomplished by de-warping the
front and back images to a set of reference markers (with known
position) that are imaged in conjunction with the body part.
However, it is desirable that the images be acquired without the
use of any reference markers to preclude the possibility of
obscuration of the important diagnostic regions of the image. It is
therefore desirable to develop an image processing algorithm that
can automatically (1) conduct image demagnification, (2) correct
the translation and rotational displacements between the front and
back images, and (3) make use of the information in the front and
back images to form a composite full image that has high geometric
fidelity without relying on any reference markers.
SUMMARY OF THE INVENTION
[0006] According to the present invention, there is provided a
solution to the problems discussed above.
[0007] According to a feature of the present invention there is
provided a method of forming a composite image from first and
second digital images formed by recording first and second
contiguous segments of a larger radiographic image in first and
second overlapping storage phosphor members, exposed to a source of
X-rays wherein the image content in the overlapped region is the
same in both images and the end edge of said first member is
present both on said first image and as a shadow edge in said
second image, said method comprising: correcting for geometric
distortion in said first and second digital images; determining any
rotational displacement and any vertical displacement between said
first and second images by matching said first member end edge in
said first image to its shadow in said second image; correcting for
image orientation based on any said rotational displacement;
determining any horizontal displacement between said first and
second images by correlating said image content in said overlapped
region of said first and second images; and stitching said first
and second images together along said first member end edge based
on any said horizontal and vertical displacements.
ADVANTAGEOUS EFFECT OF THE INVENTION
[0008] The invention has the following advantages.
[0009] 1. Enables the generation of a full composite image from two
partial images that is free from artifacts by completely
eliminating the use of references.
[0010] 2. Preserves a high degree of geometric accuracy in the
stitched image.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIGS. 1A and 1B are diagrammatic views showing an extended
length CR cassette with two storage phosphor screens built
inside.
[0012] FIG. 2 is a perspective view showing one storage phosphor
screen pulled from one end of the cassette as it would be during
processing in CR reader. The other end of the cassette is capable
of opening in a like manner.
[0013] FIG. 3 is a diagrammatic view showing two phosphor screens
which partially overlap in the center of the cassette. The
deflectors guide the screens as they approach the center of the
cassette to force the overlap.
[0014] FIGS. 4A, 4B and 4C are diagrammatic views respectively
showing how the extended length cassette is used to acquire images,
how an object of rectangular shape placed in the patient location
is deformed by magnification due to distance variation from the
x-ray source to the storage phosphor screen, and how the acquired
front and back image look. The CR reader over-scans both phosphor
screens in the vertical direction of the cassette in order to make
the screen ending edges fully visible in both images.
[0015] FIG. 5 is a flow diagram showing the image processing steps
for automatic formation of a full composite image from first and
second images.
[0016] FIG. 6 shows the major image processing steps that are used
to automatically find the locations and orientations of the screen
ending edges in both the front and the back images, and for finding
the location and orientation of the shadow of the front screen
ending edge in the back image.
[0017] FIG. 7 shows the major image processing steps that are used
for finding the horizontal displacement between the front and back
images by image-correlation.
[0018] FIG. 8 shows the composite, stitched full image.
DETAILED DESCRIPTION OF THE INVENTION
[0019] In general, the present invention relates to the
radiographic imaging of an elongate object such as the full spine
(for diagnosing scoliosis, for example) or leg of a human
subject.
[0020] Two contiguous CR plates contained in an elongated cassette
are exposed to a radiographic image of an elongate object to
produce a latent image stored in the CR plates. The CR plates are
removably mounted in the cassette and are sequentially fed to a CR
reader where the latent radiographic images are converted to two
electronic images which are combined to form an elongated image.
The elongate image can be displayed on an electronic display or
printed out on hard copy media.
[0021] Referring now to FIGS. 1A and 1B, there is shown an
embodiment of the present invention. As shown, storage phosphor
cassette 10 includes an elongate rectangular shell 12 having first
and second open ends 14 and 16. A first storage phosphor plate
assembly 18 is detachably mounted in shell 12 from the first open
end 14. A second storage phosphor plate assembly 20 is detachably
mounted on shell 12 from the second open end 16. Each assembly 18,
20 includes a respective storage phosphor plate 22, 24 and a
support and latching assembly 26, 28. Plates 22, 24 are butt joined
or overlapped in the central region 29 of shell 12. Shell 12
includes upper and lower members 30, 32 and side extrusions 34, 36
which together form a rectangular shell.
[0022] FIG. 2 shows first storage phosphor assembly 18 partially
detached from cassette 10 at a reading device (not shown).
[0023] FIG. 3 shows a cross-section of cassette 10 showing upper
and lower members 30, 32 having respective opposed inner surfaces
40, 42 including deflectors 44, 46 extending therefrom for guiding
the inner ends of assemblies 18, 20 to overlap. This results in an
overlapping of storage phosphor plates 22 and 24 to form a
composite storage phosphor plate for elongate radiographic images,
such as the human spine and leg. A more detailed description of the
latching and unlatching system and CR reader is given in U.S.
patent application Ser. No.: 09/742,509 filed Dec. 20, 2000, the
contents of which are incorporated herein by reference.
[0024] The first and second images read from first and second
storage phosphor plates 22 and 24 are formed into a composite image
according to the method of the present invention as follows.
[0025] An overview illustration of the steps involved in the
present invention is shown in FIGS. 4-9. The generation of a full
composite image from the front and back images is comprised of the
following steps: (1) demagnification of each image pixel based on
the distance between the x-ray source and the physical location of
the pixel in the individual phosphor screen, (2) determination of
the rotational displacement and the vertical displacement between
the front and back images by matching the front screen ending edge
in the front image to it's shadow in the back image, (3) image
orientation correction based on the rotational displacement, (4)
determination of the horizontal displacement between the front and
back images by correlating the image information in the overlapping
screen regions, and (5) stitching the front and the back images
together along the front screen ending edge based on the horizontal
and vertical displacements.
[0026] As shown in FIG. 4A, during an x-ray exam, the patient 403
is positioned in the path of the x-ray beam 402 from the x-ray tube
401. The extended length cassette 405 is placed behind the patient
in order to record the image of the patient. The extended length
cassette 405 can be used with an anti-scatter grid 404, which is
positioned behind the patient 403 but directly in front of the
cassette. The grid can be either a stationary type or moving bucky.
After the x-ray generator is fired and the cassette is exposed, the
image of the patient is recorded by the front screen 406 and the
back screen 407 that are enclosed inside the cassette. Each screen
captures only a portion of the image of the patient, as indicated
by element 420 and 421 (FIG. 4C). Because the front screen 406 is
not totally opaque to the incident x-rays, the back screen 407 can
still record the image of the patient 403 in the screen overlap
region 427. However, the signal-to-noise ratio of the image
captured on the back screen in the overlap region 427 will be
relatively low because of the x-ray attenuation caused by the front
screen . The image content recorded by the two screens in the
overlap region is the same. This redundant information is then used
to register the front and back images to produce a full patient
image. The front screen ending edge 408 can impose a distinctive
edge shadow on the back screen. By comparing the location and
orientation of the front screen ending edge with its shadow in the
back image, the relative orientation and vertical displacement
between the two images is determined. The exposure process
described in this paragraph corresponds to element 500 in FIG.
5.
[0027] After the x-ray exposure, the extended length CR cassette is
sent to the CR reader for image readout. The front phosphor screen
410 is scanned using a laser beam in a line-by-line format as
described by element 412. The depicted signal from the phosphor
screen, which is linearly proportional to the magnitude of the
recorded patient image signal, is extracted and converted into
digital format. The CR reader may stop the reading process when the
laser scan line nearly reaches the screen ending edge 414. This
does not guarantee that the complete information of the ending edge
will be recorded in the acquired image, which is required by this
invention for image registration. To address this issue, the CR
reader must over-scan the phosphor screen, i.e., scan slightly
beyond the end of the screen. In FIG. 4C, element 420 represents
the image acquired from the front screen, and shows that the screen
ending edge, 422, is captured completely inside the image. The
front image therefore is partitioned into two regions by the screen
ending edge: the normal image area and the over-scanned image area.
After the front screen is scanned and the image is stored in the CR
memory, the cassette is removed from the reader and inverted to
allow the back screen 407 (411) to be read in the same manner as
the front screen. However, because of the inversion of the
cassette, the laser beam conducts the scan in a format as indicated
by element 413. Therefore, to restore the correct orientation of
the back image, the acquired image must be flipped once
horizontally and once vertically after being stored in the CR
memory. Element 421 shows the acquired back image after the flip
operations. Since both the back and front screens are of the same
size, the back screen will also be over-scanned beyond its ending
edge 415 (FIG. 4B). Consequently, the screen ending edge 423 will
be captured completely inside the acquired back image 421. Due to
screen overlap, the front screen ending edge 414 is also recorded
by the back screen, which is indicated by element 425. The back
image is therefore partitioned into three regions by the shadow of
the front screen ending edge and further by the back screen ending
edge. The end-to-end readout and storage process described in this
paragraph corresponds to element 502, 504, 503, 505, and 506 as
shown in FIG. 5.
[0028] Because the front and back storage phosphor screens are not
exactly co-planar inside the extended cassette, there is a location
dependent, although slight, geometric distortion (magnification)
that is introduced, as indicated by elements 416 and 417. For a
nominal SID (x-ray source distance) of 180 cm, the mismatch between
the front and back images in the overlap region can be as large as
0.5 mm in the image horizontal axis. This can significantly impact
the stitching precision and introduce discontinuity adjacent to the
seam line in the stitched image. It is therefore necessary to
perform distortion correction, especially as the distortion
conspicuity increases as the SID decreases. The distortion
correction process is accomplished using a mathematical model that
is based on the geometric placement of the phosphor screens inside
the cassette. The design of the extended length cassette forces the
top of the front screen and the bottom of the back screen to be
coplanar such that there is no geometric distortion near the two
ends of the cassette. As the distance from either cassette end to
the center decreases, the magnification increases. This phenomenon
is more dominant in the horizontal axis than that in the vertical
axis. To correct for this distortion, each pixel in the front image
is dewarped using the following equations:
x'=x,
y'=(y-y.sub.c).times.g.sub.f.times.x/x.sub.max+y.sub.c, (1)
[0029] where x and y are image pixel coordinates in the vertical
and horizontal axes, respectively, and x' and y' are the new image
pixel coordinates, respectively, g.sub.f>=1 is a constant
specific to the front image and specific to the distance from the
x-ray tube to the cassette, x.sub.max is the pixel coordinate
maximum in the vertical axis, y.sub.c is the center coordinate in
the horizontal axis of the image. The origin of the image pixel
coordinate is defined at the image upper-left corner, with the
downward-pointing vertical axis being the positive x-axis and the
right-pointing horizontal axis being the y-axis. Eq. 1 essentially
conducts variable correction for each image row but ignores the
very small distortion in the vertical direction. The correction is
conducted symmetric to the middle column of the image, which is
valid because during the x-ray exposure the central x-ray beam is
normally centered with the cassette. Similarly, the back image can
be corrected using the formula given by:
x'=x,
y'=(y-y.sub.c).times.g.sub.b.times.(x.sub.max-x)/x.sub.max+y.sub.c,
(2)
[0030] where g.sub.b<=1 is a constant specific to the back image
and specific to the distance from the x-ray tube to the cassette.
This image demagnification process is indicated by elements 507 and
508 in FIG. 5. This processing step can be ignored when the SID
becomes large (>>180 cm), as the distortion introduced by the
magnification factor is negligible.
[0031] In order to calculate the parameters that are used for
stitching the front and back images, the screen ending edges in
both the front and the back images must be located. This operation
is shown by elements 509 and 510 in FIG. 5. The pixel values in the
image region that is beyond the screen ending edge reflect the
baseline noise level of the CR reader. This is because there is no
signal contribution from the phosphor screen. Consequently, the
pixel values in these regions are relatively low in comparison to
those in the normally exposed image regions, therefore there is an
abrupt pixel value decrement/discontinuity across the screen ending
edge in the image. This pixel value discontinuity is used to detect
the location and orientation of the screen ending edges, which can
be accomplished in many ways. In the preferred embodiment of the
present invention, the detection is carried out by (1) computing
all the significant edge transition pixels in the proximity of the
screen ending edge location, and (2) performing line delineation of
the screen ending edge pixels.
[0032] Using the front image as an example, FIG. 6 describes the
preferred embodiment of the detection process. First, a narrow band
602 is extracted from the end of the front image 600. Depending on
how the phosphor screen is being scanned in the CR reader, the
orientation of the screen ending edge 601 can have a variation of
several degrees in the acquired image from one scan to the next
scan. Therefore, the size of the narrow band must be large enough
such that the entire screen ending edge is reliably extracted. For
an image that has a width of 2,048 pixels, the size of the narrow
band should be approximately 200.times.2,048 pixels.
[0033] Second, the one-dimensional derivative of the image which is
computed in the vertical direction using an operator [-1, 0, 1]. A
one-dimensional derivative operator is preferred because the pixel
value discontinuity only occurs across the edge direction, which is
always nearly horizontal, and because of the computational
efficiency advantages. A predefined threshold is used to select
only those candidate edge transition pixels which are of greater
magnitude and of falling slope. Element 603 shows the results from
this step.
[0034] Third, a linear function is fitted to the candidate edge
pixels and the best fitting parameters are obtained when the least
square error is reached. Element 604 shows the fitted linear
function overlaid on top of the edge transition pixels. The fitting
parameters describe the ending edge location and orientation:
x=k.sub.f.times.y+a.sub.f, (3)
[0035] where k.sub.f and a.sub.f are the fitting parameters with
k.sub.f the orientation and after the offset of the front screen
ending edge in the front image. Similarly, this process is
conducted for the back image 610, except rising edge transition
pixels are searched instead inside a narrow band 614 at the
beginning of the processed back image. A new function is obtained
by least-square-error fit:
x=k.sub.b.times.y+a.sub.b, (4)
[0036] where k.sub.b and a.sub.b are the fitting parameters with
k.sub.b the orientation and a.sub.b the offset of the back screen
ending edge in the back image.
[0037] Once the screen ending edge location is successfully found
in the front image, it is compared with its shadow in the back
image for image registration. To locate the shadow of the front
screen ending edge in the back image (FIG. 5, element 511), a
similar approach to element 509 is used. This is possible because
the pixel values in the back image also undergo a strong signal
intensity decrement in the screen overlap region 427 (FIG. 4C) due
to the high attenuation of the incident x-rays by the front screen
during the x-ray exposure. In order to locate the shadow of the
front screen ending edge, the location of the narrow band needs to
be defined in the back image. This can be calculated based on the
size of the overlap regions (D in mm), which is determined by
cassette design, the image pixel size (psize in mm), and the
average location of the identified back screen ending. The distance
from the center of the narrow band to the beginning of the back
image is given by:
d=D/psize+(k.sub.b.times.y.sub.c+a.sub.b). (5)
[0038] The function that is obtained using the least-square-error
fit to describe the shadow of the front screen ending edge in the
back image can be depicted as:
x=k.times.y+a, (6)
[0039] where k and a are the fitting parameters with k the
orientation and a the offset.
[0040] Theoretically, parameters k.sub.f and k should be equal
because they both represent the orientation of the front screen
ending edge. However, they may differ by as much as several degrees
in practice for several reasons such as misalignment between the
two phosphor screens in the cassette or screen positioning
variations in the CR reader during the readout process. The
deviation between k.sub.f and k represents the orientation
misalignment between the front and back images. To assure a
seamless composite image after stitching, and to preserve high
geometric fidelity, this misalignment must be corrected.
[0041] Misalignment correction is accomplished in one of three
ways: (1) rotating the front image by .theta.=a tan(k)-a
tan(k.sub.f) while keeping the back image unchanged, (2) rotating
the back image by .theta.=a tan(k.sub.f)-a tan(k) while keeping the
front image unchanged, or (3) rotating the front and back image
by-a tan(k.sub.f) and the back image by-a tan(k), respectively. The
first and the second methods have the advantage of reduced
computation because only one of the two images must be rotated.
However, the orientation of the resultant front screen ending edge,
which is also the orientation of the seam line in the composite
stitched image, may still contain some residual mis-registration in
the horizontal direction which can cause the seam line in the
stitched image to appear jagged. The third method overcomes this
disadvantage. FIG. 5, element 513 shows the effect of rotating the
back image. Element 512, which shows the effect of rotating the
front image, is optional depending on whether method 2 or method 3
was used Since the parameters that are used for aligning the front
and back images, e.g., k.sub.a, k.sub.b, k, a.sub.a, a.sub.b, and
a, are calculated before image rotation, the parameters must be
transformed accordingly to reflect the new values in the rotated
image(s). The parameters are modified by placing Eq. 3, 4, and 6
into the transform given by:
x'=x cos .theta.+y sin .theta.,
y'=-x sin .theta.+y cos .theta., (7)
[0042] where (x', y') are the new coordinates in the rotated image,
and .theta. is the rotation angle. For the simplicity of the
description, the symbols k.sub.a, k.sub.b, k, a.sub.a, a.sub.b and
a will be used to represent the new transformed values.
[0043] The vertical displacement between the front and back image,
x_offset, is defined as the vertical distance from each pixel in
the back image to origin of the front image and is given by:
x_offset=a.sub.f-a.sub.b. (8)
[0044] Using the vertical displacement guarantees that the front
and the back images are stitched along the ending edge of the front
screen. This process is indicated by element 530 in FIG. 5.
[0045] Once the back screen ending edge, as described by k.sub.b
and a.sub.b, and the shadow of the front screen ending edge, as
described by k.sub.a and a.sub.b, are successfully identified, the
location of the screen overlap region 427 (FIG. 4C) in the back
image can be defined. The screen overlap region in the back image
is located between the back screen ending edge and the shadow of
the front screen ending edge. The size of the region is calculated
based on the equation given by:
overlap_size=(k.times.y.sub.c+a)-(k.sub.b.times.y.sub.c+a.sub.b),
(9)
[0046] and the vertical displacement from the back image origin
is:
overlap_offset.sub.b=(k.sub.b.times.y.sub.c+a.sub.b). (10)
[0047] Element 515 shows the aforementioned process. Using the
computed value of overlap_size, the corresponding region in the
front image is derived. This is the region of the same size but
with a vertical displacement from the image origin defined by:
overlap_offset.sub.f=x.sub.max-(k.sub.a.times.y.sub.c+a.sub.a)-overlap.sub-
.--size. (11)
[0048] This process is suggested by element 514. After the screen
overlap regions are extracted from each image, as shown by elements
516 and 517, they are compared in the next step to find the
horizontal displacement between the front and back images.
[0049] The image content recorded in the overlap regions are the
same except for some horizontal displacement, y_offset, between the
corresponding pixels. A one-dimensional correlation function is
computed to find the displacement using the formula given by
c(.DELTA.)=.SIGMA..sub.ijF(x.sub.i, y.sub.j).times.B(x.sub.i,
y.sub.j+.DELTA.), (12)
[0050] where F(x.sub.i, y.sub.j) and B(x.sub.i, y.sub.s) is the
pixel value at (x.sub.i, y.sub.j) in the extracted overlap region
from the front and back images, respectively, and .DELTA. is the
horizontal displacement parameter for correlation. The .DELTA.
value at which c(.DELTA.) reaches a maximum is the optimal value
for y_offset.
[0051] FIG. 7 describes the preferred implementation of this
operation. First, the overlap region 702 and 703 are extracted from
the front and back images respectively. Second, element 704 is
obtained by extracting a portion of 702, then is correlated with
703 to create the correlation function c(.DELTA.), 706. Similar
results can be achieved by correlating a portion of 703 with 702.
Third, the maximum of function c(.DELTA.) is searched and the
corresponding value of .DELTA. is identified as y_offset, 707.
Because the edge information in 702 and 703, including skin line,
tissue boundaries, bone edges, collimation boundaries, and hardware
labels etc, contribute the most useful information to the
correlation, the low frequency content is removed from 702 and 703
in order to improve the correlation robustness. Normally the
correlation function is smooth, as indicated by element 810 (FIG.
8). However, if stationary grid lines are present in the image,
small periodic peaks can appear in the function, as indicated by
element 811 (FIG. 8). The stationary grid imposes a periodic line
pattern artifact in the acquired images, the artifact is
particularly dominant when the grid is orientated in the vertical
direction, and can correlate with itself, causing small spikes to
be introduced on top of the back ground correlation function. This
artifact will negatively impact the accuracy of the determination
of the location of the true function maximum. To address this
issue, low-pass filtering of the correlation function is used
before searching for the maximum. The process described in this
paragraph is represented by element 531 (FIG. 5).
[0052] After both the front and the images have been demagnified,
the relative orientation of the two images has been aligned, and
x_offset and y_offset have been found, the back image is stitched
to the front image. Each pixel of the front image is copied to the
stitched image buffer except those pixels that are beyond the
screen ending edge line. Each pixel in the back image is copied to
the stitched image buffer with an displacement defined by x offset
and y offset except those pixels before the shadow of the front
screen ending edge. The resultant image is shown in FIG. 9. The
process conducted in this paragraph is represented by element 532
(FIG. 5).
[0053] The invention has been described in detail with particular
reference to certain preferred embodiments thereof, but it will be
understood that variations and modifications can be effected within
the spirit and scope of the invention.
Parts List
[0054] 10 storage phosphor cassette
[0055] 12 elongated rectangular shell
[0056] 14 first open end
[0057] 16 second open end
[0058] 18 first phosphor plate assembly
[0059] 20 second phosphor plate assembly
[0060] 22, 24 storage phosphor plate
[0061] 26, 28 latching assembly
[0062] 29 central region
[0063] 30, 32 upper and lower members
[0064] 34, 36 side extrusions
[0065] 40, 42 inner surfaces
[0066] 44, 46 deflectors
[0067] 401 x-ray tube
[0068] 402 x-ray beam
[0069] 403 patient
[0070] 404 x-ray anti-scatter grid
[0071] 405 extended length cassette
[0072] 406 front screen (lateral view)
[0073] 407 back screen (lateral view)
[0074] 408 ending edge of front screen (lateral view)
[0075] 410 front screen (front view)
[0076] 411 back screen (front view)
[0077] 412 indication of laser scan direction for front screen
[0078] 413 indication of laser scan direction for back screen
[0079] 414 front screen ending edge (front view)
[0080] 415 back screen ending edge (front view)
[0081] 416 deformation of a rectangular object in patient position
in front screen
[0082] 417 deformation of a rectangular object in patient position
in back screen
[0083] 418 acquired image from front screen--front image
[0084] 419 acquired image from back screen--back image
[0085] 422 front screen ending edge
[0086] 423 back screen ending edge
[0087] 425 shadow of front screen ending edge in the back image
[0088] 427 screen overlap region in the back image
[0089] 500 exposure patient
[0090] 502 read front image in the CR reader
[0091] 503 read back image in the CR reader
[0092] 504 store front image in the CR reader
[0093] 505 store back image in the CR reader
[0094] 506 flip back image once in horizontal direction and once in
vertical direction
[0095] 507 optional image demagnification of the front image
[0096] 508 optional image demagnification of the back image
[0097] 509 detect back screen ending edge in back image
[0098] 510 detect front screen ending edge in front image
[0099] 511 detect shadow of front screen ending edge in back
image
[0100] 512 optional rotation of front image
[0101] 513 image rotation of back image
[0102] 514 define screen overlap region in front image
[0103] 515 define screen overlap region in back image
[0104] 516 extract screen overlap region from front image
[0105] 517 extract screen overlap region from back image
[0106] 530 calculate vertical displacement--x_offset
[0107] 531 calculate horizontal displacement--y_offset by image
correlation
[0108] 532 image stitching
[0109] 600 acquired front image
[0110] 601 front screen ending edge
[0111] 602 extracted narrow band at the end of front image for
identifying screen ending edge
[0112] 603 candidate edge transition pixels (falling slope) in
602
[0113] 604 fitted line overlaid on top of candidate edge transition
pixels
[0114] 610 acquired back image
[0115] 611 back screen ending edge
[0116] 612 shadow of front screen ending edge in the back image
[0117] 614 extracted narrow band at the beginning of back image for
identifying screen ending edge
[0118] 615 candidate edge transition pixels (rising slope) in
614
[0119] 616 fitted line overlaid on top of candidate edge transition
pixels
[0120] 622 extracted narrow band for searching of shadow of front
screen ending edge
[0121] 623 candidate edge transition pixels (rising edge) in
622
[0122] 624 fitted line overlaid on top of candidate edge transition
pixels
[0123] 702 extract screen overlap region from front image
[0124] 703 extracted screen overlap region from back image
[0125] 704 portion of 702
[0126] 705 process for conducting image correlation
[0127] 706 correlation function
[0128] 707 location of maximum in the correlation function
[0129] 810 low-pass filtered 811
[0130] 811 spikes in the correlation function due to the use of
grid
* * * * *