Method of database-guided segmentation of anatomical structures having complex appearances
First Claim
Patent Images
1. A method for segmenting an organ comprising the steps of:
- detecting an organ with a trained detector, wherein the step of detecting an organ comprises;
receiving a first image that includes an organ; and
applying a detector trained on a database of images of like organs to the first image to discriminate the organ in the first image from a background of the age; and
determining a most likely shape of the organ, wherein the determining step comprises;
identifying images from the database of images of like organs that are closest to the organ in the first image by using a distance between learned image features that best characterize the shape of the organ,wherein the database of images of like organs is used to learn the image features that best characterize the shape of the organ by emulating a distance between feature vectors to a distance between shapes;
inferring the shape of the organ in the first image by using the identified database images to first determine the whole shape of the organ and then determine smaller parts of the shape; and
segmenting the organ from the first image by using the inferred shape of the organ,wherein the method is performed using a processor, andwherein d(cq,cr)=(cq−
cr)T(cq−
cr) is the distance between shapes, d(fq,fr)=(fq−
fr)TΣ
(fq−
fr) is the distance between feature vectors, where f is an appearance feature vector, c is its corresponding shape, (fq,cq),(fr,cr) represent the vector of a query and a reference, respectively, and Σ
is a linear metric associated with the feature vector space.
4 Assignments
0 Petitions
Accused Products
Abstract
A method for segmenting an anatomical structure of interest within an image is disclosed. The anatomical structure of interest is compared to a database of images of like anatomical structures. Those database images of like anatomical structures that are similar to the anatomical structure of interest are identified. The identified database images are used to detect the anatomical structure of interest in the image. The identified database images are also used to determine the shape of the anatomical structure of interest. The anatomical structure of interest is segmented from the image.
-
Citations
18 Claims
-
1. A method for segmenting an organ comprising the steps of:
-
detecting an organ with a trained detector, wherein the step of detecting an organ comprises; receiving a first image that includes an organ; and applying a detector trained on a database of images of like organs to the first image to discriminate the organ in the first image from a background of the age; and determining a most likely shape of the organ, wherein the determining step comprises; identifying images from the database of images of like organs that are closest to the organ in the first image by using a distance between learned image features that best characterize the shape of the organ, wherein the database of images of like organs is used to learn the image features that best characterize the shape of the organ by emulating a distance between feature vectors to a distance between shapes; inferring the shape of the organ in the first image by using the identified database images to first determine the whole shape of the organ and then determine smaller parts of the shape; and segmenting the organ from the first image by using the inferred shape of the organ, wherein the method is performed using a processor, and wherein d(cq,cr)=(cq−
cr)T(cq−
cr) is the distance between shapes, d(fq,fr)=(fq−
fr)TΣ
(fq−
fr) is the distance between feature vectors, where f is an appearance feature vector, c is its corresponding shape, (fq,cq),(fr,cr) represent the vector of a query and a reference, respectively, and Σ
is a linear metric associated with the feature vector space. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A method for segmenting an organ comprising the steps of:
-
detecting an organ with a trained detector, wherein the step of detecting an organ comprises; receiving a first image that includes an organ; and applying a detector trained on a database of images of like organs to the first image to discriminate the organ in the first image from a background of the first image; and determining a most likely shape of the organ, wherein the detector is trained by; using a weighted alignment scheme to increase influence of stable landmark points associated with the organ in the first image; eliminating influences of known invalid regions in the first image by using an occlusion mask; and using a boosted cascade of simple classifiers to train the detector on valid regions of the first image, wherein with a weight matrix W, a minimized criterion for aligning shapes is given by ℑ
GPA=∥
siRici+ti−
c ∥
w, where ci represents the ith shape control point, si,Ri,ti represent scale, rotation and translation, respectively, andc is the mean shape, andwherein the method is performed using a processor.
-
-
18. A method for segmenting an organ comprising the steps of:
-
detecting an organ with a trained detector, wherein the step of detecting an organ comprises; receiving a first image that includes an organ; and applying a detector trained on a database of images of like organs to the first image to discriminate the organ in the first image from a background of the first image; and determining a most likely shape of the organ, wherein the detector is trained by; using a weighted alignment scheme to increase influence of stable landmark points associated with the organ in the first image; eliminating influences of known invalid regions in the first image by using an occlusion mask; and using a boosted cascade of simple classifiers to train the detector on valid regions of the first image, wherein eliminating influences of known invalid regions in the first image by using an occlusion mask comprises; determining, an intensity value I for each pixel (x, y) in the first image and, for each pixel at a location (x0, y0) in the first image, computing its intensity value as follows;
-
Specification