Method and system for coronary artery centerline extraction

1. A method of extracting at least one coronary artery centerline from a 3D volume, comprising:

segmenting heart chambers in the 3D volume;

initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chambers;

locally refining points of the coronary artery centerline;

shrinking a length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery; and

extending the coronary artery centerline using vessel tracing after shrinking the length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery.

2. The method of claim 1 , wherein the step of segmenting heart chambers in the 3D volume comprises:

segmenting the heart chambers in the 3D volume using marginal space learning (MSL) based segmentation.

3. The method of claim 1 , wherein the step of initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models in a learned mean shape model.

4. The method of claim 3 , wherein the step of determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers based on a relative position of a mean coronary artery centerline model to mean heart chamber models in a learned mean shape model comprises:

calculating a deformation field from the learned mean shape model to the 3D volume based on the segmented heart chambers; and

warping the mean coronary artery centerline model from the learned mean shape model to the 3D volume using the deformation field.

5. The method of claim 3 , wherein the at least one coronary artery comprises a left main (LM) coronary artery, a left anterior descending (LAD) coronary artery, a left circumflex (LCX) coronary artery, and a right coronary artery (RCA), and the mean shape model includes:

a mean LM coronary artery model;

a mean LAD coronary artery model of a truncated portion of the LAD coronary artery based on a relative length of the LAD coronary artery and a distance between a left coronary ostium and a left ventricular apex in each of a plurality of training volumes;

a mean LCX coronary artery model of a truncated portion of the LCX coronary artery based on a relative length of the LCX coronary artery and an atrio-ventricular groove in each of the plurality of training volumes; and

a mean RCA coronary artery model of a truncated portion of the RCA coronary artery based on a relative length of the RCA coronary artery and the atrio-ventricular groove in each of the plurality of training volumes.

6. The method of claim 1 , further comprising detecting at least one coronary landmark in the 3D volume, wherein the step of initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers and the detected at least one coronary landmark based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models and the at least one coronary landmark in a learned mean shape model.

7. The method of claim 6 , wherein the step of determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers and the detected at least one coronary landmark based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models and the at least one coronary landmark in a learned mean shape model comprises:

calculating a deformation field from the learned mean shape model to the 3D volume based on a plurality of mesh points in the segmented heart chambers and the detected at least one coronary landmark; and

warping the mean coronary artery centerline model from the learned mean shape model to the 3D volume using the deformation field.

8. The method of claim 7 , wherein the at least one coronary artery comprises a left main (LM) coronary artery, a left anterior descending (LAD) coronary artery, a left circumflex (LCX) coronary artery, and a right coronary artery (RCA); the at least one coronary artery landmark comprises a left coronary ostium, a right coronary ostium, and an LM bifurcation; and the step of calculating a deformation field from the learned mean shape model to the 3D volume based on a plurality of mesh points in the segmented heart chambers and the at detected least one coronary landmark comprises:

calculating a deformation field to warp an LM coronary artery centerline model and an LAD coronary artery centerline model based on the left coronary ostium, the LM bifurcation, and a plurality of mesh points closest to the LM and LAD coronary artery centerline models in the mean shape model;

calculating a deformation field to warp an LCX coronary artery centerline model based on the LM bifurcation and a plurality of mesh points closest to the LCX coronary artery centerline model in the mean shape model; and

calculating a deformation field to warp an RCA coronary artery centerline model based on the right coronary ostium and a plurality of mesh points closest to the RCA coronary artery centerline model in the mean shape model.

9. The method of claim 1 , wherein the step of locally refining points of the coronary artery centerline comprises:

generating candidate positions for each of the points of the coronary artery centerline in a plane perpendicular to the coronary artery centerline; and

selecting the best candidate position for each of the points of the coronary artery centerline using a dynamic programming based shortest path computation based on a vesselness measure at each candidate position.

10. The method of claim 1 , wherein the step of shrinking a length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery comprises:

removing end points from the coronary artery centerline until a vesselness score for a certain number of end points is above a threshold.

11. The method of claim 10 , wherein the step of removing end points from the coronary artery centerline until a vesselness score for a certain number of end points is above a threshold comprises:

removing end points from the coronary artery until three consecutive end points have a vesselness score that is above the threshold.

12. The method of claim 1 , wherein the step of extending the coronary artery centerline using vessel tracing comprises:

(a) determining a plurality of possible paths of a particular length from a current end point of the coronary artery centerline;

(b) selecting a best path of the plurality of possible paths using a shortest path computation based on a vesselness measure; and

(c) repeating steps (a) and (b) until it is determined that a distal end of the at least one coronary artery is reached.

13. The method of claim 12 , wherein step (c) comprises:

determining a cost of the selected best path based on the vesselness measure at each point in the selected best path;

if the cost of the selected best path is less than a threshold, determining that the distal end of the at least one coronary artery is not reached and repeating steps (a) and (b); and

if the cost of the selected best path is greater than a threshold, determining that the distal end of the at least one coronary artery is reached.

14. An apparatus for extracting at least one coronary artery centerline from a 3D volume, comprising:

means for segmenting heart chambers in the 3D volume;

means for initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chambers;

means for locally refining points of the coronary artery centerline;

means for shrinking a length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery; and

means for extending the coronary artery centerline using vessel tracing after shrinking the length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery.

15. The apparatus of claim 14 , wherein the means for initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

means for determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models in a learned mean shape model.

16. The apparatus of claim 15 , wherein the at least one coronary artery comprises a left main (LM) coronary artery, a left anterior descending (LAD) coronary artery, a left circumflex (LCX) coronary artery, and a right coronary artery (RCA), and the mean shape model includes:

a mean LM coronary artery model;

a mean LAD coronary artery model of a truncated portion of the LAD coronary artery based on a relative length of the LAD coronary artery and a distance between a left coronary ostium and a left ventricular apex in each of a plurality of training volumes;

a mean LCX coronary artery model of a truncated portion of the LCX coronary artery based on a relative length of the LCX coronary artery and an atrio-ventricular groove in each of the plurality of training volumes; and

a mean RCA coronary artery model of a truncated portion of the RCA coronary artery based on a relative length of the RCA coronary artery and the atrio-ventricular groove in each of the plurality of training volumes.

17. The apparatus of claim 14 , further comprising means for detecting at least one coronary landmark in the 3D volume, wherein the means for initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

means for determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers and the detected at least one coronary landmark based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models and the at least one coronary landmark in a learned mean shape model.

18. The apparatus of claim 14 , wherein the means for locally refining points of the coronary artery centerline comprises:

means for generating candidate positions for each of the points of the coronary artery centerline; and

means for selecting the best candidate position for each of the points of the coronary artery centerline.

19. The apparatus of claim 14 , wherein the means for extending the coronary artery centerline using vessel tracing comprises:

means for selecting a best path of the plurality of possible paths using a shortest path computation based on a vesselness measure; and

means for determining whether a distal end of the at least one coronary artery is reached.

20. A non-transitory computer readable medium storing computer program instructions, which when executed on a processor, cause the processor to perform a method of extracting at least one coronary artery centerline from a 3D volume, the method comprising:

segmenting heart chambers in the 3D volume;

initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chambers;

locally refining points of the coronary artery centerline;

shrinking a length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery; and

extending the coronary artery centerline using vessel tracing after shrinking the length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery.

21. The non-transitory computer readable medium of claim 20 , wherein the step of initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models in a learned mean shape model.

22. The non-transitory computer readable medium of claim 21 , wherein the at least one coronary artery comprises a left main (LM) coronary artery, a left anterior descending (LAD) coronary artery, a left circumflex (LCX) coronary artery, and a right coronary artery (RCA), and the mean shape model includes:

a mean LM coronary artery model;

a mean LAD coronary artery model of a truncated portion of the LAD coronary artery based on a relative length of the LAD coronary artery and a distance between a left coronary ostium and a left ventricular apex in each of a plurality of training volumes;

a mean LCX coronary artery model of a truncated portion of the LCX coronary artery based on a relative length of the LCX coronary artery and an atrio-ventricular groove in each of the plurality of training volumes; and

a mean RCA coronary artery model of a truncated portion of the RCA coronary artery based on a relative length of the RCA coronary artery and the atrio-ventricular groove in each of the plurality of training volumes.

23. The non-transitory computer readable medium of claim 20 , wherein the method further comprises detecting at least one coronary landmark in the 3D volume, and the step of initializing a coronary artery centerline in the 3D volume for at least a portion of at least one coronary artery based on the segmented heart chamber comprises:

determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers and the detected at least one coronary landmark based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models and the at least one coronary landmark in a learned mean shape model.

24. The non-transitory computer readable medium of claim 23 , wherein the step of determining an initial position of the coronary artery centerline in the 3D volume from the segmented heart chambers and the detected at least one coronary landmark based on a relative position of a mean coronary artery centerline model for the at least one coronary artery to mean heart chamber models and the at least one coronary landmark in a learned mean shape model comprises:

calculating a deformation field from the learned mean shape model to the 3D volume based on a plurality of mesh points in the segmented heart chambers and the detected at least one coronary landmark; and

warping the mean coronary artery centerline model from the learned mean shape model to the 3D volume using the deformation field.

25. The non-transitory computer readable medium of claim 24 , wherein the at least one coronary artery comprises a left main (LM) coronary artery, a left anterior descending (LAD) coronary artery, a left circumflex (LCX) coronary artery, and a right coronary artery (RCA); the at least one coronary artery landmark comprises a left coronary ostium, a right coronary ostium, and an LM bifurcation; and the step of calculating a deformation field from the learned mean shape model to the 3D volume based on a plurality of mesh points in the segmented heart chambers and the at detected least one coronary landmark comprises:

calculating a deformation field to warp an LM coronary artery centerline model and an LAD coronary artery centerline model based on the left coronary ostium, the LM bifurcation, and a plurality of mesh points closest to the LM and LAD coronary artery centerline models in the mean shape model;

calculating a deformation field to warp an LCX coronary artery centerline model based on the LM bifurcation and a plurality of mesh points closest to the LCX coronary artery centerline model in the mean shape model; and

calculating a deformation field to warp an RCA coronary artery centerline model based on the right coronary ostium and a plurality of mesh points closest to the RCA coronary artery centerline model in the mean shape model.

26. The non-transitory computer readable medium of claim 20 , wherein the step of locally refining points of the coronary artery centerline comprises:

generating candidate positions for each of the points of the coronary artery centerline in a plane perpendicular to the coronary artery centerline; and

selecting the best candidate position for each of the points of the coronary artery centerline using a dynamic programming based shortest path computation based on a vesselness measure at each candidate position.

27. The non-transitory computer readable medium of claim 20 , wherein the step of shrinking a length of the coronary artery centerline to verify that the coronary artery centerline is within the at least one coronary artery comprises:

removing end points from the coronary artery centerline until a vesselness score for a certain number of end points is above a threshold.

28. The non-transitory computer readable medium of claim 20 , wherein the step of extending the coronary artery centerline using vessel tracing comprises:

(a) determining a plurality of possible paths of a particular length from a current end point of the coronary artery centerline;

(b) selecting a best path of the plurality of possible paths using a shortest path computation based on a vesselness measure; and

(c) repeating steps (a) and (b) until it is determined that a distal end of the at least one coronary artery is reached.

29. The non-transitory computer readable medium of claim 28 , wherein step (c) comprises:

determining a cost of the selected best path based on the vesselness measure at each point in the selected best path;

if the cost of the selected best path is less than a threshold, determining that the distal end of the at least one coronary artery is not reached and repeating steps (a) and (b); and

if the cost of the selected best path is greater than a threshold, determining that the distal end of the at least one coronary artery is reached.