Learning-based aorta segmentation using an adaptive detach and merge algorithm

Patent metadata

Field	Value
Publication number	US-9589211-B2
Application number	US-201514707503-A
Country	US
Kind code	B2
Filing date	May 8, 2015
Priority date	May 8, 2015
Publication date	Mar 7, 2017
Grant date	Mar 7, 2017

Abstract

Systems and methods for segmenting a structure of interest in medical imaging data include generating a binary mask highlighting structures in medical imaging data, the highlighted structures comprising a connected component including a structure of interest. A probability map is computed by classifying voxels in the highlighted structures using a trained classifier. A plurality of detaching operations is performed on the highlighted structures to split the connected component into a plurality of detached connected components. An optimal detaching parameter is determined representing a number of the detaching operations. A detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter is classified as the structure of interest based on the probability map and the trained classifier.

First claim

The invention claimed is:

1. A method for segmenting a structure of interest in medical imaging data, comprising:
generating a binary mask highlighting structures in medical imaging data, the highlighted structures comprising a connected component including a structure of interest;
computing a probability map by classifying voxels in the highlighted structures using a trained classifier;
performing a plurality of detaching operations on the highlighted structures to split the connected component into a plurality of detached connected components;
determining an optimal detaching parameter representing a number of the detaching operations; and
classifying a detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter as the structure of interest based on the probability map and the trained classifier.

2. The method as recited in claim 1 , wherein generating a binary mask highlighting structures in medical imaging data comprises:
applying intensity thresholding to the medical imaging data to determine the highlighted structures.

3. The method as recited in claim 1 , wherein computing a probability map by classifying voxels in the highlighted structures using a trained classifier comprises at least one of:
computing a probability of each voxel in the highlighted structures as being a bone using a trained bone classifier; and
computing a probability of each voxel in the highlighted structures as being a vessel using a trained vessel classifier.

4. The method as recited in claim 1 , wherein performing a plurality of detaching operations on the highlighted structures to split the connected component into a plurality of detached connected components comprises:
performing a plurality of erosion operations to remove a layer of voxels from a periphery of the highlighted structures.

5. The method as recited in claim 1 , wherein determining an optimal detaching parameter representing a number of the detaching operations comprises:
determining the optimal detaching parameter representing the number of the detaching operations that provides for a maximum information gain.

6. The method as recited in claim 1 , wherein determining an optimal detaching parameter representing a number of the detaching operations comprises:
generating a tree graph for the connected component in the highlighted structures representing how the connected component splits into the plurality of detached connected components over the plurality of detaching operations.

7. The method as recited in claim 6 , wherein determining an optimal detaching parameter representing a number of the detaching operations further comprises:
calculating a probability for each node in the tree graph based on the probability map; and
determining the optimal detaching parameter based on the probability for each node in the tree graph.

8. The method as recited in claim 6 , wherein generating a tree graph for the connected component in the highlighted structures representing how the connected component splits into the plurality of detached connected components over the plurality of detaching operations comprises:
representing the plurality of detached connected components as respective child nodes in the tree graph.

9. The method as recited in claim 1 , wherein determining an optimal detaching parameter representing a number of the detaching operations comprises:
after each of the plurality of detaching operations:
labeling each detached connected component resulting from a respective one of the plurality of detaching operations based on probabilities associated with voxels in the respective detached connected component; and
dilating each labeled detached connected component.

10. The method as recited in claim 1 , wherein classifying a detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter as the structure of interest based on the probability map comprises:
computing an average probability of voxels of the detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter as being one of a bone and a vessel; and
comparing the average probability with a threshold to classify the detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter as the structure of interest.

11. The method as recited in claim 1 , wherein performing a plurality of detaching operations on the highlighted structures to split the connected component into a plurality of detached connected components comprises:
determining top N largest connected components in the highlighted structures, where N is any positive integer; and
performing the plurality of detaching operations on the top N largest connected components.

12. The method as recited in claim 1 , wherein generating a binary mask highlighting structures in medical imaging data comprises:
generating the binary mask highlighting high contrast structures in a contrast enhanced three-dimension computed tomography (CT) volume.

13. An apparatus for segmenting a structure of interest in medical imaging data, comprising:
means for generating a binary mask highlighting structures in medical imaging data, the highlighted structures comprising a connected component including a structure of interest;
means for computing a probability map by classifying voxels in the highlighted structures using a trained classifier;
means for performing a plurality of detaching operations on the highlighted structures to split the connected component into a plurality of detached connected components;
means for determining an optimal detaching parameter representing a number of the detaching operations; and
means for classifying a detached connected component resulting from performing the number of detaching operations corresponding to the optimal detaching parameter as the structure of interest based on the probability map and the trained classifier.

14. The apparatus as recited in claim 13 , wherein the means for generating a binary mask highlighting structures in medical imaging data comprises:
means for applying intensity thresholding to the medical imaging data to determine the highlighted structures.

15. The apparatus as recited in claim 13 , wherein the means for computing a probability map by classifying voxels in the highlighted structures using a trained classifier comprises at least one of:
means for computing a probability of each voxel in the highlighted structures as being a bone using a trained bone classifier; and
means for computing a probability of each voxel in the highlighted structures as being a vessel using a trained vessel classifier.

16. The apparatus as recited in claim 13 , wherein the means for performing a plurality of detaching operations on the highlighted structures to split the connected component into a plurality of detached connected components comprises:
means for performing a plurality of erosion operations to remove a layer of voxels from a periphery of the highlighted structures.

17. The apparatus as recited in claim 13 , wherein the means for determining an optimal detaching parameter representing a number of the detaching operations comprises:
means for determining the optimal detaching parameter representing the number of the detaching operations that provides for a maximum information gain.

18. The apparatus as recited in claim

Assignees

• Siemens Healthcare Gmbh

Inventors

• Lay Nathan
• Liu David
• Kretschmer Jan
• Zhou Shaohua Kevin

Classifications

• G06V10/764 Primary

  using classification, e.g. of video objects · CPC title

• G06V10/457

  by analysing connectivity, e.g. edge linking, connected component analysis or slices · CPC title

• G06V10/426

  Graphical representations · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06T11/26

  Drawing of charts or graphs · CPC title