Method and system for multi-part left atrium segmentation in C-arm computed tomography volumes using shape constraints

1. A method of segmenting a left atrium (LA) in a 3D volume comprising:

segmenting an LA chamber body mesh, an appendage mesh, and a plurality of pulmonary vein (PV) meshes in the 3D volume, wherein the LA chamber body mesh and the appendage mesh are segmented as a combined object and the plurality of PV meshes are segmented subject to a statistical shape constraint, and the segmenting of the plurality of PV meshes subject to the statistical shape constraint comprises:

independently estimating pose parameters for each of the plurality of PV meshes in the 3D volume using marginal space learning (MSL),

generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes, and

adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model; and

generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes.

2. The method of claim 1 , wherein the plurality of PV meshes includes a left inferior PV mesh, a left superior PV mesh, a right inferior PV mesh, and a right superior PV mesh.

3. The method of claim 1 , wherein the segmenting of the LA chamber body mesh and the appendage mesh as a combined object comprises:

segmenting the combined object including the LA chamber body mesh and the appendage mesh using marginal space learning (MSL).

4. The method of claim 1 , wherein the step of generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes comprises:

for each of the plurality of PV meshes, converting the pose parameters estimated for the PV mesh to a set of four-point set representing the estimated pose parameters, wherein the four-point set includes a center point and three corner points of a bounding box defined by the estimated pose parameters for the PV mesh; and

generating the point distribution model including each of the four-point sets representing the estimated pose parameters of the plurality of PV meshes, the center point of the segmented chamber body mesh, and the center point of the segmented appendage mesh.

5. The method of claim 4 , wherein the step of adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model comprises:

deforming the point distribution model using an active shape model to enforce the statistical shape constraint;

determining an adjusted pose for each of the plurality of PV meshes based on the deformed point distribution model.

6. The method of claim 1 , wherein the step of generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes comprises:

connecting the LA chamber mesh with the appendage mesh and each of the plurality of PV meshes by projecting the appendage mesh and each of the plurality of PV meshes to the LA chamber mesh, resulting in a connected set of meshes;

converting the connected set of meshes to a volume mask; and

generating the consolidated LA mesh from the volume mask.

7. The method of claim 6 , wherein the step of generating the consolidated LA mesh from the volume mask comprises:

generating the consolidated LA mesh from the volume mask using a marching cubes algorithm.

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

9. An apparatus for segmenting a left atrium (LA) in a 3D volume comprising:

means for segmenting an LA chamber body mesh, an appendage mesh, and a plurality of pulmonary vein (PV) meshes in the 3D volume, wherein the means for segmenting an LA chamber body mesh, an appendage mesh, and a plurality of pulmonary vein (PV) meshes in the 3D volume comprises means for segmenting the LA chamber body mesh and the appendage mesh as a combined object and means for segmenting the plurality of PV meshes subject to a statistical shape constraint, and the means for segmenting the plurality of PV meshes subject to the statistical shape constraint comprises:

means for independently estimating pose parameters for each of the plurality of PV meshes in the 3D volume using marginal space learning (MSL),

means for generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes, and

means for adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model; and

means for generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes.

10. The apparatus of claim 9 , wherein the plurality of PV meshes includes a left inferior PV mesh, a left superior PV mesh, a right inferior PV mesh, and a right superior PV mesh.

11. The apparatus of claim 9 , wherein the means for segmenting the LA chamber body mesh and the appendage mesh as a combined object comprises:

means for segmenting the combined object including the LA chamber body mesh and the appendage mesh using marginal space learning (MSL).

12. The apparatus of claim 9 , wherein the means for generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes comprises:

means for converting the pose parameters estimated for a PV mesh to a set of four-point set representing the estimated pose parameters, wherein the four-point set includes a center point and three corner points of a bounding box defined by the estimated pose parameters for the PV mesh; and

means for generating the point distribution model including the four-point sets representing the estimated pose parameters of each of the plurality of PV meshes, the center point of the segmented chamber body mesh, and the center point of the segmented appendage mesh.

13. The apparatus of claim 12 , wherein the means for adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model comprises:

means for deforming the point distribution model using an active shape model to enforce the statistical shape constraint;

means for determining an adjusted pose for each of the plurality of PV meshes based on the deformed point distribution model.

14. The apparatus of claim 9 , wherein the means for generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes comprises:

means for connecting the LA chamber mesh with the appendage mesh and each of the plurality of PV meshes to result in a connected set of meshes;

means for converting the connected set of meshes to a volume mask; and

means for generating the consolidated LA mesh from the volume mask.

15. A non-transitory computer readable medium encoded with computer executable instructions for segmenting a left atrium (LA) in a 3D volume, the computer executable instructions defining a method comprising:

segmenting an LA chamber body mesh, an appendage mesh, and a plurality of pulmonary vein (PV) meshes in the 3D volume, wherein the LA chamber body mesh and the appendage mesh are segmented as a combined object and the plurality of PV meshes are segmented subject to a statistical shape constraint, and the segmenting of the plurality of PV meshes subject to the statistical shape constraint comprises:

independently estimating pose parameters for each of the plurality of PV meshes in the 3D volume using marginal space learning (MSL),

generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes, and

adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model; and

generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes.

16. The non-transitory computer readable medium of claim 15 , wherein the plurality of PV meshes includes a left inferior PV mesh, a left superior PV mesh, a right inferior PV mesh, and a right superior PV mesh.

17. The non-transitory computer readable medium of claim 15 , wherein the segmenting of the LA chamber body mesh and the appendage mesh as a combined object comprises:

segmenting the combined object including the LA chamber body mesh and the appendage mesh using marginal space learning (MSL).

18. The non-transitory computer readable medium of claim 15 , wherein the step of generating a point distribution model based on the pose parameters and center points of the segmented LA chamber body and appendage meshes comprises:

for each of the plurality of PV meshes, converting the pose parameters estimated for the PV mesh to a set of four-point set representing the estimated pose parameters, wherein the four-point set includes a center point and three corner points of a bounding box defined by the estimated pose parameters for the PV mesh; and

generating the point distribution model including each of the four-point sets representing the estimated pose parameters of the plurality of PV meshes, the center point of the segmented chamber body mesh, and the center point of the segmented appendage mesh.

19. The non-transitory computer readable medium of claim 18 , wherein the step of adjusting the pose parameters for the plurality of PV meshes by enforcing a statistical shape constraint on the generated point distribution model comprises:

deforming the point distribution model using an active shape model to enforce the statistical shape constraint;

determining an adjusted pose for each of the plurality of PV meshes based on the deformed point distribution model.

20. The non-transitory computer readable medium of claim 15 , wherein the step of generating a consolidated LA mesh from the LA chamber body mesh, the appendage mesh and the plurality of PV meshes comprises:

connecting the LA chamber mesh with the appendage mesh and each of the plurality of PV meshes by projecting the appendage mesh and each of the plurality of PV meshes to the LA chamber mesh, resulting in a connected set of meshes;

converting the connected set of meshes to a volume mask; and

generating the consolidated LA mesh from the volume mask.

21. The non-transitory computer readable medium of claim 20 , wherein the step of generating the consolidated LA mesh from the volume mask comprises:

generating the consolidated LA mesh from the volume mask using a marching cubes algorithm.