Method and system for automatic aorta segmentation

Patent metadata

Field	Value
Publication number	US-9715637-B2
Application number	US-72567910-A
Country	US
Kind code	B2
Filing date	Mar 17, 2010
Priority date	Mar 18, 2009
Publication date	Jul 25, 2017
Grant date	Jul 25, 2017

Abstract

A method and system for aorta segmentation in a 3D volume, such as a C-arm CT volume is disclosed. The aortic root is detected in the 3D volume using marginal space learning (MSL) based segmentation. The aortic arch is detected in the 3D volume using MSL based segmentation. The ascending aorta is tracked from the aortic root to the aortic arch in the 3D volume, and the descending aorta is tracked from the aortic arch in the 3D volume.

First claim

The invention claimed is:

1. A method for aorta segmentation in a 3D volume, comprising:
detecting an aortic root in the 3D volume using marginal space learning (MSL); and
tracking an ascending aorta from the detected aortic root in the 3D volume; and
generating a segmented aorta including the detected aortic root and the ascending aorta.

2. The method of claim 1 , wherein the 3d volume is a C-arm CT volume.

3. The method of claim 1 , wherein said step of detecting an aortic root in the 3D volume using marginal space learning (MSL) comprises:
detecting a position, orientation, and scale of the aortic root in the 3D volume using a trained position classifier, a trained position-orientation classifier, and a trained position-orientation-scale classifier;
aligning a mean truncated aortic root estimated from a set of training volumes to the detected position, orientation, and scale in the 3D volume to generate an initial detection result for the aortic root; and
refining a boundary of the initial detection result for the aortic root using a trained boundary classifier.

4. The method of claim 3 , wherein the mean truncated aortic root is estimated by identifying a portion of the aortic root that is common to all of the training volumes, and truncating an annotated aortic root in each of the training volumes to match the portion of the aortic root that is common to all training volumes.

5. The method of claim 1 , wherein said step of tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
detecting, on a slice by slice basis starting at the aortic root and moving upward in the 3D volume, an aortic circle representing an intersection of the ascending aorta and a current slice using a trained 2D circle detector.

6. The method of claim 1 , further comprising:
detecting an aortic arch in the 3D volume using MSL.

7. The method of claim 6 , wherein said step of tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
tracking the ascending aorta from the detected aortic root to the detected aortic arch.

8. The method of claim 7 , further comprising:
tracking a descending aorta from the detected aortic arch, wherein the segmented aorta includes the aortic root, the ascending aorta, the aortic arch, and the descending aorta.

9. The method of claim 8 , wherein said step of tracking a descending aorta from the detected aortic arch comprises:
detecting, on a slice by slice basis starting at the aortic arch and moving downward in the 3D volume, an aortic circle representing an intersection of the descending aorta and a current slice using a trained 2D circle detector; and
stopping said tracking at a slice in which no aortic circle is detected by the 2D circle detector.

10. The method of claim 6 , wherein said step of detecting an aortic arch in the 3D volume using MSL comprises:
detecting a position, orientation, and scale of the aortic arch in the 3D volume using a trained position classifier, a trained position-orientation classifier, and a trained position-orientation-scale classifier;
aligning a mean truncated aortic arch estimated from a set of training volumes to the detected position, orientation, and scale in the 3D volume to generate an initial detection result for the aortic arch; and
refining a boundary of the initial detection result for the aortic arch using a trained boundary classifier.

11. The method of claim 10 , wherein the mean truncated aortic arch is estimated by identifying portion of the aortic arch that is common to all of the training volumes, and truncating an annotated aortic arch in each of the training volumes to match the portion of the aortic arch that is common to all training volumes.

12. The method of claim 1 , further comprising determining that no aortic arch is present in the 3D volume, wherein said step of tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
detecting, on a slice by slice basis starting at the aortic root and moving upward in the 3D volume, an aortic circle representing an intersection of the ascending aorta and a current slice using a trained 2D circle detector; and
stopping said tracking at one of a slice in which no aortic circle is detected by the 2D circle detector and a top border of the 3D volume.

13. The method of claim 1 , further comprising:
refining a boundary of the segmented aorta using trained boundary detector to adjust each point on the boundary of the segmented aorta.

14. The method of claim 13 , further comprising:
smoothing the refined boundary of the segmented aorta.

15. A method for aorta segmentation in a 3D volume, comprising:
detecting an aortic root in the 3D volume using marginal space learning (MSL);
detecting an aortic arch in the 3D volume using MSL if the aortic arch is present in the 3D volume;
if the aortic arch is not detected in the 3D volume:
tracking an ascending aorta upward from the detected aortic root in the 3D volume;

if the aortic arch is detected in the 3D volume:
tracking the ascending aorta from the detected aortic root to the detected aortic arch in the 3D volume, and
tracking a descending aorta from the detected aortic arch in the 3D volume.

16. An apparatus for aorta segmentation in a 3D volume, comprising:
means for detecting an aortic root in the 3D volume using marginal space learning (MSL); and
means for tracking an ascending aorta from the detected aortic root in the 3D volume; and
means for generating a segmented aorta including the detected aortic root and the ascending aorta.

17. The apparatus of claim 16 , wherein said means for tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
means for detecting, on a slice by slice basis starting at the aortic root and moving upward in the 3D volume, an aortic circle representing an intersection of the ascending aorta and a current slice using a trained 2D circle detector.

18. The apparatus of claim 16 , further comprising:
means for detecting an aortic arch in the 3D volume using MSL; and
means for tracking a descending aorta from the detected aortic arch, wherein the segmented aorta includes the aortic root, the ascending aorta, the aortic arch, and the descending aorta.

19. The apparatus of claim 18 , wherein said means for tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
means for tracking the ascending aorta from the detected aortic root to the detected aortic arch.

20. The apparatus of claim 19 , wherein said means for tracking a descending aorta from the detected aortic arch comprises:
means for detecting, on a slice by slice basis starting at the aortic arch and moving downward in the 3D volume, an aortic circle representing an intersection of the

descending aorta and a current slice using a trained 2D circle detector; and
means for stopping said tracking at a slice in which no aortic circle is detected by the 2D circle detector.

21. The apparatus of claim 16 , further comprising:
means for determining that no aortic arch is present in the 3D volume;
wherein said step of tracking an ascending aorta from the detected aortic root in the 3D volume comprises:
means for detecting, on a slice by slice basis starting at the aortic root and moving upward in the 3D volume, an aortic circle representing an intersection of the ascending aorta and a current slice using a trained 2D circle detector; and
means for stopping said tracking a

Assignees

• Zheng Yefeng
• Georgescu Bogdan
• John Matthias
• Boese Jan
• Comaniciu Dorin
• Siemens Healthcare Gmbh

Inventors

• Zheng Yefeng
• Georgescu Bogdan
• John Matthias
• Boese Jan
• Comaniciu Dorin

Classifications

• G06V10/457

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

• G06T7/12 Primary

  Edge-based segmentation · CPC title

• G06K9/4638 Primary

  Physics · mapped topic

• G06T2207/30101

  Blood vessel; Artery; Vein; Vascular · CPC title

• G06T2200/04

  involving 3D image data · CPC title