Coronary reconstruction from rotational X-ray projection sequence

1. A computer readable medium embodying instructions executable by a processor to perform a method for three-dimensional reconstruction of a branched object from a rotational sequence of images of the branched object, the method steps comprising:

generating a vesselness feature image for each image of the sequence by filtering each image of the sequence using a filter that identifies tubular structure;

segmenting the branched object from each vesselness feature image using a thresholding process;

extracting centerlines of the branched object;

performing symbolic reconstruction via a stereo correspondence matching between the centerlines from different views of the sequence of images using a graph cut-based optimization; and

creating a three-dimensional tomographic reconstruction of the branched object compensated for motion of the branched object between the images of the sequence.

2. The computer readable medium of claim 1 , wherein the graph cut-based optimization comprises:

constructing a connectivity graph for the centerlines extracted from a reference image selected from among the sequence of images;

locating an optical center in the reference image;

projecting a three-dimensional ray from optical center to the extracted centerlines in the reference image;

tracking the three-dimensional ray in a volume constructed from the sequence of images and labeling points on the ray at different depths according to a depth in the volume;

projecting the points onto another image according to a respective projection matrices; and

minimizing a mean of errors between the projected points and corresponding points on the centerlines extracted from other images.

3. The computer readable medium of claim 2 , wherein the graph cut-based optimization further comprises a smoothness penalty for neighboring connected pixels based on a difference in depth.

4. The computer readable medium of claim 2 , wherein the graph cut-based optimization further comprises a cost penalty increasing with length of a corresponding centerline.

5. The computer readable medium of claim 2 , wherein the graph cut-based optimization further comprises a cost penalty decreasing with a vesselness of a corresponding centerline.

6. The computer readable medium of claim 1 , wherein creating the three-dimensional tomographic reconstruction further comprises estimating an affine transformation for each image of the sequence to compensate for the motion in a two-dimensional imaging plane.

7. The computer readable medium of claim 1 , wherein a Multiscale Hessian-based filter is used in the filtering of each image of the sequence.

8. The computer readable medium of claim 1 , wherein the thresholding process is a hysteresis thresholding process that retains connected components for points that have a vesselness above a low threshold and with at least one pixel with a vesselness above a high threshold.

9. The computer readable medium of claim 8 , wherein the low and high thresholds are calculated as fixed quantities of accumulated histograms of the vesselness feature images.

10. A computer-implemented method for three-dimensional reconstruction of a branched object from a rotational sequence of images of the branched object comprising:

generating a vesselness feature image for each image of the sequence by filtering each image of the sequence using a filter that identifies tubular structure;

segmenting the branched object from each vesselness feature image of the sequence using a thresholding process;

extracting centerlines of the branched object;

performing symbolic reconstruction via a stereo correspondence matching between the centerlines from different views of the sequence of images using a graph cut-based optimization; and

creating a three-dimensional tomographic reconstruction of the branched object compensated for motion of the branched object between the images of the sequence.

11. The method of claim 10 , wherein the graph cut-based optimization comprises:

constructing a connectivity graph for the centerlines extracted from a reference image selected from among the sequence of images;

locating an optical center in the reference image;

projecting a three-dimensional ray from optical center to the extracted centerlines in the reference image;

tracking the three-dimensional ray in a volume constructed from the sequence of images and labeling points on the ray at different depths according to a depth in the volume;

projecting the points onto another image according to a respective projection matrices; and minimizing a mean of errors between the projected points and corresponding points on the centerlines extracted from other images.

12. The method of claim 11 , wherein the graph cut-based optimization further comprises a smoothness penalty for neighboring connected pixels based on a difference in depth.

13. The method of claim 11 , wherein the graph cut-based optimization further comprises a cost penalty increasing with length of a corresponding centerline.

14. The method of claim 11 , wherein the graph cut-based optimization further comprises a cost penalty decreasing with a vesselness of a corresponding centerline.

15. The method of claim 10 , wherein creating the three-dimensional tomographic reconstruction further comprises estimating an affine transformation for each image of the sequence to compensate for the motion in a two-dimensional imaging plane.

16. The computer-implemented method of claim 10 , wherein a Multiscale Hessian-based filter is used in the filtering of each image of the sequence.

17. The computer-implemented method of claim 10 , wherein the thresholding process is a hysteresis thresholding process that retains connected components for points that have a vesselness above a low threshold and with at least one pixel with a vesselness above a high threshold.

18. A system for three-dimensional reconstruction of a branched object from a rotational sequence of images of the branched object comprising:

a source of the rotational sequence of images comprising views of the branched object to be reconstructed;

a processor executing instructions embodied in a computer readable medium for three-dimensional reconstruction of the branched object from a symbolic reconstruction of the branched object in the rotational sequence of images; and

an output device for display of the reconstructed branched object determined by the processor,

wherein the instructions cause the processor to perform method steps comprising:

generating a vesselness feature image for each image of the sequence by filtering each image of the sequence using a filter that identifies tubular structure;

segmenting the branched object from each vesselness feature image using a thresholding process;

extracting centerlines of the branched object;

creating the symbolic reconstruction of the branched object via a stereo correspondence matching between the centerlines from different views of the sequence of images; and

creating a three-dimensional tomographic reconstruction of the branched object compensated for motion of the branched object between the images of the sequence in the symbolic reconstruction.

19. The system of claim 18 , wherein a Multiscale Hessian-based filter is used in the filtering of each image of the sequence.

20. The system of claim 18 , wherein the thresholding process is a hysteresis thresholding process that retains connected components for points that have a vesselness above a low threshold and with at least one pixel with a vesselness above a high threshold.