System and method for auto-contouring in adaptive radiotherapy

1. A method, implemented by a processor, for performing auto-contouring in a medical image, the method comprising:

receiving the medical image, wherein the medical image contains a region of interest and a plurality of voxels;

identifying, by the processor, voxels of the medical image which are on a surface of the region of interest;

selecting, by the processor, a first set of the identified voxels to form a first contour;

refining the first contour, by the processor, based on a shape dictionary, to generate a second contour of the region of interest;

updating a boundary detector based on a seconds set of the identified voxels, the second set being located along the second contour; and

applying the updated boundary detector to auto-contour the region of interest in the medical image.

2. The method of claim 1 , further including refining, by the processor, the second contour using the boundary detector.

3. The method of claim 1 , wherein:

identifying voxels of the medical image which are on the surface of the region of interest includes determining a probability of the voxels being on the surface of the region of interest; and

selecting the first set of the identified voxels includes selecting voxels having probabilities higher than a predetermined threshold.

4. The method of claim 1 , wherein refining, by the processor, the first contour based on the shape dictionary to generate the second contour includes:

selecting a set of shapes from the shape dictionary;

combining the selected set of shapes to approximate the first contour; and

generating the second contour based on the combined set of shapes.

5. The method of claim 4 , further including:

determining whether at least a subset of voxels on the first contour are on the second contour by minimizing an optimization function.

6. The method of claim 1 , wherein updating the boundary detector includes:

comparing the medical image having the second contour with a collection of stored images containing the region of interest;

selecting a subset of stored images with similar image structure to the medical image having the second contour; and

updating the boundary detector based on the medical image having the second contour and the selected subset of stored images.

7. The method of claim 6 , wherein updating the boundary detector includes:

selecting voxels located on surfaces of the selected subset of stored images;

extracting steerable features from the selected voxels located on sufaces of the selected subset of stored images;

determining a feature vector based on the extracted steerable features; and

updating the boundary detector based on the feature vector.

8. The method of claim 1 , further including updating the shape dictionary based on the second contour, wherein updating the shape dictionary includes:

selecting a subset of shapes from the medical image containing the second contour;

obtaining a sparse coefficient for each shape in the subset; and

updating the shape dictionary based on the obtained sparse coefficients.

9. The method of claim 1 , wherein updating the boundary detector includes:

(a) after updating the boundary detector based on the second contour, determining an updated first contour;

(b) refining the updated first contour based on the shape dictionary to generate an updated second contour; and

repeating steps (a) and (b) until a number of repetitions reaches a predetermined value.

10. The method of claim 1 , wherein updating the boundary detector includes:

(a) after updating the boundary detector based on the second contour, determining an updated first contour;

(b) refining the updated first contour based on the shape dictionary to generate an updated second contour: and

repeating steps (a) and (b) until a residual error reaches a predetermined minimum.

11. A system for performing auto-contouring in a medical image, comprising:

a processor;

a memory operatively coupled to the processor and storing instructions, that when executed by the processor, causes the processor to perform a method, the method comprising:

receiving the medical image, wherein the medical image contains a region of interest and a plurality of voxels;

identifying voxels. of the medical image which are on a surface of the region of interest;

selecting a first set of the identified voxels to form first contour;

refining the first contour based on a shape dictionary to generate a second contour of the region of interest;

updating a boundary detector based on a second set of the identified voxels, the second set being located along the second contour; and

applying the updated boundary detector to auto contour the region of interest in the medical image.

12. The system of claim 11 , wherein identifying voxels of the medical image includes determining a probability of the voxels being on the surface; and selecting voxels having probabilities higher than a predetermined threshold.

13. The system of claim 11 , wherein refining the first contour based on the shape dictionary to generate the second contour of the region of interest includes:

selecting a set of shapes from the shape dictionary;

combining the selected set of shapes to approximate the first contour; and

generating the second contour based on the combined set of shapes.

14. The system of claim 13 , wherein the method performed by the processor further includes:

determining whether at least a subset of voxels on the first contour are on the second contour by minimizing an optimization function.

15. The system of claim 11 , wherein the method performed by the processor further includes refining the second contour using the boundary detector.

16. The system of claim 11 , wherein updating the boundary detector includes:

(a) after updating the boundary detector based on the second contour, determining an updated first contour;

(b) refining the updated first contour based on the shape dictionary to generate an updated second contour; and

repeating steps (a) and (b) until a number of repetitions reaches a predetermined value.

17. The system of claim 11 , wherein updating the boundary detector includes:

(a) after updating the boundary detector based on the second contour, determining an updated first contour:

(b) refining the updated first contour based on the shape dictionary to generate an updated second contour; and

repeating steps (a) and (b) until a residual error reaches a predetermined minimum.

18. A computer-implemented method for training a boundary detector based on a plurality of medical images containing a region of interest, the method comprising:

selecting, by a processor from the plurality of medical images, voxels located on a contour surface of the region of interest

extracting, by the processor, steerable features from the selected voxels;

determining, by the processor, a feature vector based on the extracted steerable features; and

generating, by the processor, the boundary detector based on the feature vector.

19. The computer-implemented method of claim 18 , comprising:

comparing a new image having a new contour surface with the plurality of medical images;

selecting a subset of medical images from the plurality of medical images based on image structure of the new image; and

updating the boundary detector based on the new image and the selected subset of medical images.

20. The computer-implemented method of claim 19 , further including:

selecting a subset of shapes from the new image;

obtaining sparse coefficients for shapes in the subset; and

updating a shape dictionary based on the obtained sparse coefficients.