System and method for segmenting the left ventricle in a cardiac MR image

1. A method for segmenting a magnetic resonance image of interest of a left ventricle comprising the steps of:

determining a myocardium contour according to a graph cut of candidate endocardium contours, and a spline fitting to candidate epicardium contours in the absence of shape propagation; and

applying a plurality of shape constraints to the candidate endocardium contours and the candidate epicardium contours, to determine the myocardium contour, wherein a template is determined by shape propagation of a plurality of magnetic resonance images in a sequence including the magnetic resonance image of interest, in the presence of shape propagation.

2. A method for segmenting a magnetic resonance image of interest of a left ventricle comprising the steps of:

determining a myocardium response image according to a histogram of pixel intensity in magnetic resonance image of interest;

determining a plurality of candidate contours, according to a plurality of energy functions, to which a plurality of confidence values are assigned; and

applying a plurality of shape constraints to the candidate endocardium contours and the candidate epicardium contours to determine a myocardium contour.

3. The method of claim 2 , wherein the myocardium contour is based on a plurality of magnetic resonance images in a sequence including the magnetic resonance image of interest, wherein the myocardium contour includes an endocardium contour and an epicardium contour.

4. The method of claim 2 , further comprising the step of determining an approximate contour pair of the left ventricle according to a plurality of points defined in a gray level profile of the image as an intersection of a plurality of cross sections of the left ventricle.

5. The method of claim 4 , wherein the approximate contour pair is determined according to a Hough transform array to vote for a center position and a radius of a myocardium centerline.

6. The method of claim 2 , wherein the step of determining the myocardium response image according to the histogram of pixel intensity further comprises the step of fitting a mixture of Gaussians to the histogram of pixel intensity to determine portions of blood, muscle and air in the image.

7. The method of claim 2 , wherein for each candidate closed contour, the method further comprises the steps of:

defining a search space around an approximate contour;

defining a line of source nodes and sink nodes;

determining a shortest path between the source nodes and the sink nodes;

defining a new source node and a plurality of new sink nodes; and

determining a shortest path between the new source node and new sink nodes.

8. The method of claim 2 , wherein the candidate endocardium contours and the candidate epicardium contours are each determined according to a combination of a gradient magnitude and a direction using a cross product of a gradient direction and a contour direction.

9. The method of claim 8 , further comprising the steps of:

determining a first candidate endocardium contour according to a positive cross product of a gradient direction in the input image and a contour direction;

determining a second candidate endocardium contour according to a negative cross product of a gradient direction in the myocardium image and a contour direction;

determining a first candidate epicardium contour according to a gradient magnitude in the input image; and

determining a second candidate epicardium contour according to a positive cross product of a gradient direction in the myocardium image and a contour direction.

10. The method of claim 2 , further comprising the step of determining a one-to-one match matrix, wherein points on the endocardium contour and epicardium contour are assigned to a plurality of template points, wherein the template points are determined from propagation through a plurality of magnetic resonance images.

11. A method for segmenting a magnetic resonance image of interest of a left ventricle comprising the steps of:

determining a myocardium response image according to a histogram of pixel intensity in magnetic resonance image of interest;

determining a plurality of candidate contours, according to a plurality of energy functions, to which a plurality of confidence values are assigned;

determining a myocardium contour according to a graph cut of candidate endocardium contours, and a spline fitting to candidate epicardium contours in the absence of shape propagation; and

applying a plurality of shape constraints to the candidate endocardium contours and the candidate epicardium contours to determine the myocardium contour, wherein a template is determined by shape propagation of a plurality of magnetic resonance images in a sequence including the magnetic resonance image of interest, in the presence of shape propagation.

12. The method of claim 11 , wherein the myocardium contour includes an endocardium contour and an epicardium contour.

13. The method of claim 11 , further comprising the step of determining an approximate contour pair of the left ventricle according to a plurality of points defined in a gray level profile of the image as an intersection of a plurality of cross sections of the left ventricle.

14. The method of claim 13 , wherein the approximate contour pair is determined according to a Hough transform array to vote for a center position and a radius of a myocardium centerline.

15. The method of claim 11 , wherein the step of determining the myocardium response image according to the histogram of pixel intensity further comprises the step of fitting a mixture of Gaussians to the histogram of pixel intensity to determine portions of blood, muscle and air in the image.

16. The method of claim 11 , wherein the closed contour includes a plurality of candidate contours, and for each contour, further comprises the steps of:

defining a search space around an approximate contour;

defining a line of source nodes and sink nodes;

determining a shortest path between the source nodes and the sink nodes;

defining a new source node and a plurality of new sink nodes; and

determining a shortest path between the new source node and new sink nodes.

17. The method of claim 11 , wherein the candidate endocardium contour and the candidate epicardium contour are each determined according to a combination of a gradient magnitude and a direction using a cross product of a gradient direction and a contour direction.

18. The method of claim 17 , further comprising the steps of:

determining a first candidate endocardium contour according to a positive cross product of a gradient direction in the input image and a contour direction;

determining a second candidate endocardium contour according to a negative cross product of a gradient direction in the myocardium image and a contour direction;

determining a first candidate epicardium contour according to a gradient magnitude in the input image; and

determining a second candidate epicardium contour according to a positive cross product of a gradient direction in the myocardium image and a contour direction.

19. The method of claim 11 , wherein the graph cut further includes the steps of:

defining a graph wherein each node corresponds to a connected component region between confidence pixels on the candidate contours; and

determining a minimum cut between a center node and an outside node.

20. The method of claim 11 , wherein the spline fitting further includes the step of fitting a spline through points of the candidate contours.

21. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for segmenting a magnetic resonance image of interest of a left ventricle, the method steps comprising:

determining a myocardium response image according to a histogram of pixel intensity in magnetic resonance image of interest;

determining a plurality of candidate contours, according to a plurality of energy functions, to which a plurality of confidence values are assigned;

determining a myocardium contour according to a graph cut of candidate endocardium contours, and a spline fitting to candidate epicardium contours in the absence of shape propagation; and

applying a plurality of shape constraints to the candidate endocardium contour and the candidate epicardium contours to determine the myocardium contour, wherein a template is determined by shape propagation of a plurality of magnetic resonance images in a sequence including the magnetic resonance image of interest, in the presence of shape propagation.