Computer graphics processing and selective visual display system – Computer graphics processing – Three-dimension
Reexamination Certificate
2002-02-07
2004-05-18
Zimmerman, Mark (Department: 2671)
Computer graphics processing and selective visual display system
Computer graphics processing
Three-dimension
C345S420000, C345S593000, C345S594000, C382S130000, C382S154000
Reexamination Certificate
active
06738063
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to the identification of object-correspondence between two image sets in general, and to lung nodule correspondence identification between medical image sets in particular.
BACKGROUND OF THE INVENTION
Lung CT technology has been widely used by physicians in the screening and diagnosis of lung cancer. From lung CT images, the physician can search for nodules and make judgements on their malignancy based on the statistics of the nodules, including shape, size, etc. A very important piece of information is the status change of the nodules, such as changes in shape, size, and density. One of the most significant quantitative measurements is the growth rate of lung nodules during a period of time. It is therefore crucial to identify the correspondence of the same nodule in two or more lung CT image sets captured at different time frames.
So far, this task has been done manually, and therefore it is tedious, slow, and error prone because of the tremendous amount of data. Because the CT data are 3D images, the task becomes very difficult for the physician, if at all achievable. In current clinical practice, the physician is required to scan through 2D slices of the 3D image data one by one and try to find the correspondence of a nodule in two image sets. The number of slices for a single data set is as large as several hundreds, and a single slice contains 250,000 pixels. Moreover, the imaging condition for the patient in the CT studies may be varied and the organ and the body may be deformed between two such studies. In many cases, it is difficult to determine if a nodule disappears after a period of time, or it still exists, because the physician is not able to identify the correspondence.
Fast registration of local volumes of interest (VOI) from large 3D image data is very often needed in medical image analysis systems, such as the systems for analyzing lung CT images. For example, in the screening and diagnosis of lung cancer, very important pieces of information are the presence of a new nodule, the absence of a previously presented nodule, and the growth rate of a lung nodule. It is therefore crucial to identify the correspondence of the same nodule in two or more lung CT image sets captured at different time frames. In most cases, the properties of the nodule and its surrounding structures are locally distinct, and therefore the registration of local VOI's is sufficient for identifying the correspondence of nodules.
The conventional algorithms for correspondence identification often use full volume registration/warping, which has severe shortcomings. First, the size of the data set is very large. A typical data set is 512×512×280, which makes full volume registration/warping out of the question if reasonable speed is required. Second, the poses of the patient and the lung volume controls during two image scans are always somewhat different, which causes non-linear distortions between the two result data sets. Therefore, to do an accurate full volume warping, non-linear techniques will be needed, which are complicated, difficult, slow, and unreliable.
What is needed is a system that avoids full volume registration, but performs fast and accurate registration of two local VOI's.
SUMMARY OF THE INVENTION
Disclosed is a method for object correspondence identification between two image sets, said method steps comprising receiving a selected point in said one image set, locating a rough matching point in said another image set, defining a first volume-of-interest around said selected point, defining a search window around said matching point comprising a plurality of neighboring points, for every point in said search window, defining a second volume-of-interest, computing the similarity between each said second volume-of-interest with said first volume-of-interest, and selecting that second volume-of-interest that is most similar to said first volume-of-interest.
In another aspect of the invention, said receiving of a selected point is effected through a graphical user interface.
Another aspect of the invention further comprises the step of roughly aligning the two image sets.
In another aspect of the invention, said rough alignment is effected through area and boundary matching.
In another aspect of the invention, said step of computing the similarity between said first and second volumes of interest comprises a grayscale cross correlation computation according to the equation
S
(
V
A
,
V
B
)
=
∑
(
V
A
(
x
,
y
,
z
)
-
V
_
A
)
(
V
B
(
x
,
y
,
z
)
-
V
_
B
)
∑
(
V
A
(
x
,
y
,
z
)
-
V
_
A
)
2
∑
(
V
B
(
x
,
y
,
z
)
-
V
_
B
)
2
where {overscore (V)}
A
and {overscore (V)}
B
are the mean of the grayscale values of all pixels in said two volumes-of-interest and the summations are over all the voxels in both said volumes-of-interest.
In another aspect of the invention, said step of computing the similarity between said first and second volumes of interest comprises computing the sum of absolute differences over all the voxels in said volumes-of-interest in accordance with the equation
S
(
V
A
,V
B
)=&Sgr;|
V
A
(
x,y,z
)−
V
B
(
x,y,z
)|
In another aspect of the invention, said step of computing the similarity between said first and second volumes of interest comprises thresholding followed by summation of inclusive OR:
I
(
x
,
y
,
z
)
=
{
1
,
V
A
(
x
,
y
,
z
)
≥
T
,
V
B
(
x
,
y
,
z
)
≥
T
1
,
V
A
(
x
,
y
,
z
)
<
T
,
V
B
(
x
,
y
,
z
)
<
T
,
and
0
,
else
S
(
V
A
,V
B
)=&Sgr;|
I
(
x,y,z
)|
where T is the threshold, and the summation is over all the voxels in both said volumes-of-interest.
In another aspect of the invention, said step of computing the similarity between said first and second volumes of interest comprises surface matching wherein the surface points of all objects within each volume-of-interest are defined as those points inside said volume-of-interest that have at least one immediate neighbor whose intensity is above a given threshold T and at least one immediate neighbor whose intensity is below T, and performing the steps of constructing a three-dimensional distance map for a first said volume-of-interest, V
A
, D
A
(x,y,z) such that its value is the distance of the given point (x,y,z) to the closest surface point within V
A
, and calculating the dissimilarity between said volumes-of-interest according to the equation
ds
=
∑
(
x
,
y
,
z
)
∈
s
B
D
A
(
x
,
y
,
z
)
.
where s
B
represents the complete set of surface points in said second volume-of-interest V
B
. In case the two data sets are of different resolutions, the coordinates of V
A
and V
B
need to be normalized before the above equation can be applied.
Also disclosed is a program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform the method steps set forth above.
REFERENCES:
patent: 5647360 (1997-07-01), Bani-Hashemi et al.
patent: 6154518 (2000-11-01), Gupta
patent: 6175655 (2001-01-01), George et al.
Mangin J.-F., et al. “Nonsupervised 3D Registration of PET and MRI Data Using Chamfer Matching,” Conference Record of the 1992 IEEE Nuclear Science Symposium and Medical Imaging Conference, Orlando, FL, USA, Oct. 25-31, 1992, pp. 1262-1264, vol. 2, XP010108658.
Brown L.G.: “A Survey of Image Registration Techniques,” ACM Computing Surveys, New York, NY, US, vol. 24, No. 4, Dec. 1, 1992 pp. 325-376, XP000561460.
International Search Report.
Fan Li
Qian Jianzhong
Shen Hong
Nguyen Kimbinh T.
Siemens Corporate Research Inc.
Zimmerman Mark
LandOfFree
Object-correspondence identification without full volume... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Object-correspondence identification without full volume..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Object-correspondence identification without full volume... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3238588