Hierarchical deformable model for image segmentation
View Patent ↗Described herein is a technology for facilitating deformable model-based segmentation of image data. In one implementation, the technology includes receiving training image data ( 202 ) and automatically constructing a hierarchical structure ( 204 ) based on the training image data. At least one spatially adaptive boundary detector is learned based on a node of the hierarchical structure ( 206 ).
1. A method for facilitating deformable model-based segmentation of image data, comprising:
receiving, by a computer system, training image data;
automatically constructing, by the computer system, based on the training image data, a hierarchical structure comprising multiple nodes, wherein a node represents a cluster of vertices having relatively similar characteristics; and
learning, by the computer system, at least one spatially adaptive boundary detector based on a node of the hierarchical structure
wherein the step of automatically constructing the hierarchical structure comprises performing a process including if a divergence between positive and negative samples in a node of the hierarchical structure is less than a pre-determined threshold, splitting the node by performing clustering based on selected features.
2. The method of claim 1 wherein the step of automatically constructing the hierarchical structure comprises performing an iterative process comprising:
clustering the training image data to form a node of the hierarchical structure; and
selecting a set of features for a next level of the clustering.
3. The method of claim 2 wherein the step of clustering comprises applying affinity propagation to the training image data.
4. The method of claim 2 further comprises determining an initial set of features for the clustering by determining an intensity profile along a normal of a vertex.
5. The method of claim 2 wherein the step of selecting the set of features comprises selecting the features with highest accuracies for classification.
6. The method of claim 2 wherein the step of selecting the set of features comprises applying an AdaBoost method.
7. The method of claim 2 wherein the step of selecting the set of features comprises applying a random forest-based method.
8. The method of claim 1 wherein the step of learning the at least one spatially adaptive boundary detector comprises applying a supervised learning method.
9. The method of claim 8 wherein the step of learning the at least one spatially adaptive boundary detector comprises applying an AdaBoost method.
10. The method of claim 1 wherein the step of learning the at least one spatially adaptive boundary detector comprises applying a random forest-based method.
11. The method of claim 1 wherein the step of learning the at least one spatially adaptive boundary detector comprises applying a support vector-based method.
12. The method of claim 1 wherein the hierarchical structure comprises a binary decision tree.
13. The method of claim 12 wherein the hierarchical structure comprises a clustering tree and multiple cluster node classifiers constructed at clustering leaf nodes of the clustering tree.
14. The method of claim 1 wherein the divergence comprises a Kullback-Leibler (KL) divergence.
15. The method of claim 1 further comprising:
selecting features for the clustering by randomly sampling from an extensive redundant feature pool.
16. The method of claim 1 wherein the step of splitting the node comprises performing clustering of vertices in the node.
17. The method of claim 16 wherein the step of performing the clustering comprises:
forming vertices;
constructing a feature vector for a vertex; and
performing two-center clustering of the vertices based on pair-wise similarities between the vertices.
18. The method of claim 17 wherein the step of forming the vertices comprises grouping at least one positive sample and at least one negative sample.
19. The method of claim 17 wherein the step of constructing the feature vector comprises applying a weak feature on voxels of the vertex.
20. The method of claim 1 further comprising:
clustering shape instances in the training image data; and
constructing multiple active shape models, based on the clustered shape instances, for constraining the deformable model-based segmentation.
21. A method of image segmentation, comprising:
receiving, by a computer system, an input image;
initializing, by the computer system, a deformable model; and
performing, by the computer system, deformation on vertices of the input image using a hierarchical structure with spatially adaptive boundary detectors to guide the deformable model to a boundary,
wherein said hierarchical structure is automatically constructed, by the computer system, based on training image data,
wherein a node of the hierarchical structure represents vertices that are driven in response to a same boundary detector;
wherein the step of automatically constructing the hierarchical structure comprises performing a process including if a divergence between positive and negative samples in a node of the hierarchical structure is less than a pre-determined threshold, splitting the node by performing clustering based on selected features.
22. A non-transitory computer usable medium having a computer readable program code tangibly embodied therein, said computer readable program code adapted to be executed by a processor to implement a method for facilitating deformable model-based segmentation of image data, comprising:
receiving training image data;
automatically constructing, based on the training image data, a hierarchical structure comprising multiple nodes, wherein a node represents a cluster of vertices having relatively similar characteristics; and
learning at least one spatially adaptive boundary detector based on a node of the hierarchical structure;
wherein the step of automatically constructing the hierarchical structure comprises performing a process including if a divergence between positive and negative samples in a node of the hierarchical structure is less than a pre-determined threshold, splitting the node by performing clustering based on selected features.
23. A system for facilitating deformable model-based segmentation of image data, comprising:
a memory device for storing computer readable program code; and
a processor in communication with the memory device, the processor being operative with the computer readable program code to:
receive training image data;
automatically construct, based on the training image data, a hierarchical structure comprising multiple nodes, wherein a node represents a cluster of vertices having relatively similar characteristics, wherein automatically constructing the hierarchical structure further comprises performing a process including if a divergence between positive and negative samples in a node of the hierarchical structure is less than a pre-determined threshold, splitting the node by performing clustering based on selected features; and
learn at least one spatially adaptive boundary detector based on a node of the hierarchical structure.