Method and system for segmentation of the prostate in 3D magnetic resonance images
View Patent ↗A method and system for fully automatic segmentation the prostate in multi-spectral 3D magnetic resonance (MR) image data having one or more scalar intensity values per voxel is disclosed. After intensity standardization of multi-spectral 3D MR image data, a prostate boundary is detected in the multi-spectral 3D MR image data using marginal space learning (MSL). The detected prostate boundary is refined using one or more trained boundary detectors. The detected prostate boundary can be split into patches corresponding to anatomical regions of the prostate and the detected prostate boundary can be refined using trained boundary detectors corresponding to the patches.
1. A method for prostate segmentation in multi-spectral 3D magnetic resonance (MR) image data, comprising:
detecting a prostate boundary in the multi-spectral 3D MR image data using marginal space learning (MSL), wherein the multi-spectral 3D MR image data comprises, for each voxel of a plurality of voxels, a respective vector of intensity values representing different MR channels for that voxel; and
refining the detected prostate boundary using one or more trained boundary detectors; wherein said step of detecting a prostate boundary in the multi-spectral 3D MR image data comprises,
detecting prostate center position candidates using a trained position detector;
detecting position-orientation candidates based on the prostate center position candidates using a trained position-orientation detector; and
detecting a full similarity transformation based on the position-orientation candidates using a trained full similarity transformation detector.
2. The method of claim 1 , further comprising:
standardizing voxel intensities of the multi-spectral 3D MR image data prior to said step of detecting a prostate boundary.
3. The method of claim 1 , wherein the position detector, the position-orientation detector, and the full similarity transformation detector are trained based on artificially generated training data generated by randomly perturbing real-world training data.
4. The method of claim 3 , wherein the artificially generated training data is randomly perturbed by perturbing position, orientation, scale, and skewness parameters of annotated prostates in the real-world training data using a random 12 parameter affine perturbation.
5. The method of claim 1 , wherein said step of refining the detected prostate boundary using one or more trained boundary detectors comprises:
splitting the detected prostate boundary into a plurality of patches corresponding to a plurality anatomical regions; and
refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches.
6. The method of claim 5 , wherein said step of refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches comprises:
maximizing a posterior probability of a global shape model for the prostate boundary using an active shape model; and
maximizing a posterior probability of boundary points in each of the plurality of patches of the detected prostate boundary using the corresponding one of the plurality of boundary detectors.
7. The method of claim 1 , wherein the multi-spectral 3D MR image data comprises a plurality of aligned mono-spectral MR images.
8. The method of claim 1 , further comprising:
detecting one or more anatomic landmarks associated with the prostate using one or more respective trained landmark detectors.
9. The method of claim 1 , further comprising:
segmenting one or more anatomical regions within the segmented prostate in the multi-spectral 3D MR image data using MSL constrained to the segmented prostate.
10. An apparatus for prostate segmentation in multi-spectral 3D magnetic resonance (MR) image data, comprising:
means for detecting a prostate boundary in the multi-spectral 3D MR image data using marginal space learning (MSL), wherein the multi-spectral 3D MR image data comprises, for each voxel of a plurality of voxels, a respective vector of intensity values representing different MR channels for that voxel; and
means for refining the detected prostate boundary using one or more trained boundary detectors; wherein said means for detecting a prostate boundary in the multi-spectral 3D MR image data comprises,
means for detecting prostate center position candidates using a trained position detector;
means for detecting position-orientation candidates based on the prostate center position candidates using a trained position-orientation detector; and
means for detecting a full similarity transformation based on the position-orientation candidates using a trained full similarity transformation detector.
11. The apparatus of claim 10 , further comprising:
means for standardizing voxel intensities of the multi-spectral 3D MR image data.
12. The apparatus of claim 10 , wherein the position detector, the position-orientation detector, and the full similarity transformation detector are trained based on artificially generated training data generated by randomly perturbing real-world training data.
13. The apparatus of claim 12 , wherein the artificially generated training data is randomly perturbed by perturbing position, orientation, scale, and skewness parameters of annotated prostates in the real-world training data using a random 12 parameter affine perturbation.
14. The apparatus of claim 10 , wherein said means for refining the detected prostate boundary using one or more trained boundary detectors comprises:
means for splitting the detected prostate boundary into a plurality of patches corresponding to a plurality anatomical regions; and
means for refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches.
15. The apparatus of claim 14 , wherein said means for refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches comprises:
means for maximizing a posterior probability of a global shape model for the prostate boundary using an active shape model; and
means for maximizing a posterior probability of boundary points in each of the plurality of patches of the detected prostate boundary using the corresponding one of the plurality of boundary detectors.
16. The apparatus of claim 10 , wherein the multi-spectral 3D MR image data comprises a plurality of aligned mono-spectral MR images.
17. The apparatus of claim 10 , further comprising:
means for detecting one or more anatomic landmarks associated with the prostate using one or more respective trained landmark detectors.
18. The apparatus of claim 10 , further comprising:
means for segmenting one or more anatomical regions within the segmented prostate in the multi-spectral 3D MR image data using MSL constrained to the segmented prostate.
19. A non-transitory computer readable medium encoded with computer executable instructions for prostate segmentation in multi-spectral 3D magnetic resonance (MR) image data, the computer executable instructions defining steps comprising:
detecting a prostate boundary in the multi-spectral 3D MR image data using marginal space learning (MSL), wherein the multi-spectral 3D MR image data comprises, for each voxel of a plurality of voxels, a respective vector of intensity values representing different MR channels for that voxel; and
refining the detected prostate boundary using one or more trained boundary detectors; wherein the computer executable instructions defining the step of detecting a prostate boundary in the multi-spectral 3D MR image data comprise computer executable instructions defining the steps of:
detecting prostate center position candidates using a trained position detector;
detecting position-orientation candidates based on the prostate center position candidates using a trained position-orientation detector; and
detecting a full similarity transformation based on the position-orientation candidates using a trained full similarity transformation detector.
20. The computer readable medium of claim 19 , further comprising computer executable instructions defining the step of:
standardizing voxel intensities of the multi-spectral 3D MR image data prior to said step of detecting a prostate boundary.
21. The computer readable medium of claim 19 , wherein the position detector, the position-orientation detector, and the full similarity transformation detector are trained based on artificially generated training data generated by randomly perturbing real-world training data.
22. The computer readable medium of claim 21 , wherein the artificially generated training data is randomly perturbed by perturbing position, orientation, scale, and skewness parameters of annotated prostates in the real-world training data using a random 12 parameter affine perturbation.
23. The computer readable medium of claim 19 , wherein the computer executable instructions defining the step of refining the detected prostate boundary using one or more trained boundary detectors comprise computer executable instructions defining the steps of:
splitting the detected prostate boundary into a plurality of patches corresponding to a plurality anatomical regions; and
refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches.
24. The computer readable medium of claim 23 , wherein the computer executable instructions defining the step of refining the detected prostate boundary using a plurality of trained boundary detectors corresponding to the plurality of patches comprise computer executable instructions defining the steps of:
maximizing a posterior probability of a global shape model for the prostate boundary using an active shape model; and
maximizing a posterior probability of boundary points in each of the plurality of patches of the detected prostate boundary using the corresponding one of the plurality of boundary detectors.
25. The computer readable medium of claim 19 , wherein the multi-spectral 3D MR image data comprises a plurality of aligned mono-spectral MR images.
26. The computer readable medium of claim 19 , further comprising computer executable instructions defining the step of:
detecting one or more anatomic landmarks associated with the prostate using one or more respective trained landmark detectors.
27. The computer readable medium of claim 19 , further comprising computer executable instructions defining the step of:
segmenting one or more anatomical regions within the segmented prostate in the multi-spectral 3D MR image data using MSL constrained to the segmented prostate.