Method and system for automatic coronary artery detection

1. A method for coronary artery detection in a 3D cardiac volume, comprising:

segmenting at least one heart chamber in said volume;

generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber,

the initial estimation of the coronary artery comprising a plurality of control points defining the coronary artery; and detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume; wherein said step of detecting the coronary artery in said 3D cardiac volume comprises: (a) detecting an optimal point within a neighborhood of each control point having a highest probability using a trained coronary artery detector; and (b) replacing each control point with the detected optimal point in the neighborhood of that control point.

2. The method of claim 1 , further comprising:

growing the detected coronary artery using 3D dynamic programming.

3. The method of claim 1 , wherein said step of segmenting at least one heart chamber comprises:

estimating a location, orientation, and scale of the heart in said volume using marginal space learning (MSL);

aligning a mean heart model to said volume based on the estimated location, orientation, and scale, said mean heart model including at least one coronary artery; and

generating a segmented heart model by deforming the mean heart model using learning based chamber boundary detectors to fit heart chamber boundaries in said volume.

4. The method of claim 3 , wherein said step of generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber comprises:

estimating a deformation field that warps the mean heart model to the segmented heart model; and

applying the estimated deformation field to the at least one coronary artery in the segmented heart model to generate the initial estimation of the coronary artery in said volume.

5. The method of claim 4 , wherein:

said step of estimating a deformation field comprises modeling the deformation field using a thin-plate spline (TPS); and

said step of applying the estimated deformation field to the at least one coronary artery in the segmented heart model comprises applying the TPS to the at least one coronary artery in the segmented heart model.

6. The method of claim 1 , wherein said step of generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber comprises:

applying at least one of a rigid transformation and a non-rigid transformation to a mean shape model of the coronary artery in said volume to deform the mean shape model of the coronary artery to generate the initial estimation of the coronary artery.

7. The method of claim 1 , wherein said trained coronary artery detector is trained based on annotated training data using a probabilistic boosting tree and local scale and shift invariant steerable features.

8. The method of claim 1 , wherein step (a) comprises: for each control point, locally searching for the point with maximum probability within a plane normal to a tangent direction of the coronary artery at the control point.

9. The method of claim 1 , wherein said step of detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume further comprises: (c) smoothing the coronary artery defined by the refined control points.

10. The method of claim 9 , wherein said step of detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume further comprises:

repeating steps (a), (b), and (c) until each control point converges.

11. An apparatus for coronary artery detection in a 3D cardiac volume, comprising: means for segmenting at least one heart chamber in said volume; means for generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber, the initial estimation of the coronary artery comprising a plurality of control points defining the coronary artery; and means for detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume; wherein said means for detecting the coronary artery comprises: means for detecting an optimal point within a neighborhood of each control point having a highest probability using a trained coronary artery detector; and means for replacing each control point with the detected optimal point in the neighborhood of that control point.

12. The apparatus of claim 11 , further comprising:

means for growing the detected coronary artery using 3D dynamic programming.

13. The apparatus of claim 11 , wherein said means for segmenting at least one heart chamber comprises:

means for estimating a location, orientation, and scale of the heart in said volume using marginal space learning (MSL);

means for aligning a mean heart model to said volume based on the estimated location, orientation, and scale, said mean heart model including at least one coronary artery; and

means for generating a segmented heart model by deforming the mean heart model using learning based chamber boundary detectors to fit heart chamber boundaries in said volume.

14. The apparatus of claim 13 , wherein said means for generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber comprises:

means for estimating a deformation field that warps the mean heart model to the segmented heart model; and

means for applying the estimated deformation field to the at least one coronary artery in the segmented heart model to generate the initial estimation of the coronary artery in said volume.

15. The apparatus of claim 11 , wherein said trained coronary artery detector is trained based on annotated training data using a probabilistic boosting tree and local scale and shift invariant steerable features.

16. The apparatus of claim 11 , wherein said means for detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume further comprises: means for smoothing the coronary artery defined by the refined control points.

17. A non-transitory computer readable medium encoded with computer executable instructions for coronary artery detection in a 3D cardiac volume, the computer executable instructions defining steps comprising: segmenting at least one heart chamber in said volume; generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber, the initial estimation of the coronary artery comprising a plurality of control points defining the coronary artery; and detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume; wherein detecting the coronary artery in said 3D cardiac volume comprise computer executable instructions defining the steps of: (a) detecting an optimal point within a neighborhood of each control point having a highest probability using a trained coronary artery detector; and (b) replacing each control point with the detected optimal point in the neighborhood of that control point.

18. The non-transitory computer readable medium of claim 17 , further comprising computer executable instructions defining the step of:

growing the detected coronary artery using 3D dynamic programming.

19. The non-transitory computer readable medium of claim 17 , wherein the computer executable instructions defining the step of segmenting at least one heart chamber comprise computer executable instructions defining the steps of:

estimating a location, orientation, and scale of the heart in said volume using marginal space learning (MSL);

aligning a mean heart model to said volume based on the estimated location, orientation, and scale, said mean heart model including at least one coronary artery; and

generating a segmented heart model by deforming the mean heart model using learning based chamber boundary detectors to fit heart chamber boundaries in said volume.

20. The non-transitory computer readable medium of claim 19 , wherein the computer executable instructions defining the step of generating an initial estimation of a coronary artery in said volume based on the segmented at least one heart chamber comprise computer executable instructions defining the steps of:

estimating a deformation field that warps the mean heart model to the segmented heart model; and

applying the estimated deformation field to the at least one coronary artery in the segmented heart model to generate the initial estimation of the coronary artery in said volume.

21. The non-transitory computer readable medium of claim 17 , wherein said trained coronary artery detector is trained based on annotated training data using a probabilistic boosting tree and local scale and shift invariant steerable features.

22. The non-transitory computer readable medium of claim 17 , wherein the computer executable instructions defining the step of detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume further comprise computer executable instructions defining the step of: (c) smoothing the coronary artery defined by the refined control points.

23. The non-transitory computer readable medium of claim 22 , wherein the computer executable instructions defining the step of detecting the coronary artery in said 3D cardiac volume by refining each of the control points in the initial estimation of the coronary artery based on local information to each of the control points in said volume further comprise computer executable instructions defining the step of:

repeating steps (a), (b), and (c) until each control point converges.