Method for mass candidate detection and segmentation in digital mammograms
View Patent ↗A basic component of Computer-Aided Detection systems for digital mammography comprises generating candidate mass locations suitable for further analysis. A component is described that relies on filtering either the background image or the complementary foreground mammographic detail by a purely signal processing method on the one hand or a processing method based on a physical model on the other hand. The different steps of the signal processing approach consist of band-pass filtering the image by one or more band pass filters, multidimensional clustering, iso-contouring of the distance to centroid of the one or more filtered values, and finally candidate generation and segmentation by contour processing. The physics-based approach also filters the image to retrieve a fat-corrected image to model the background of the breast, and the resulting image is subjected to a blob detection filter to model the intensity bumps on the foreground component of the breast that are associated with mass candidates.
1. A method of detecting candidate masses in a mammographic image, comprising:
subjecting the mammographic signal representing said image to a cascaded filtering process, wherein the cascaded filtering process comprises: 1) background removal filtering for removing a globally varying component in said mammographic image which is based on a physical model of breast composition and/or breast compression and includes a minimum tracking filter from thorax side to mammographic skin line side in which minimum filter values of the minimum tracking filter spread out in a radial way until the breast skin line is reached, and 2) foreground candidate mass filtering, and
subjecting a result of said cascaded filtering to segmentation.
2. A method according to claim 1 comprising foreground candidate mass filtering by applying at least one even symmetric Gabor filter or at least one bandpass Gabor filter.
3. A method according to claim 1 wherein said foreground candidate mass filtering is performed using a Gaussian Laplacian Pyramid image decomposition method.
4. A method according to claim 1 wherein said foreground candidate mass filtering is performed using wavelet image decomposition.
5. A method according to claim 1 comprising foreground candidate mass filtering by a blob detection process at multiple scales.
6. A method according to claim 1 wherein said background represents a fat component of a breast.
7. A method according to claim 1 wherein the foreground candidate mass filtering comprises bandpass filtering.
8. A method according to claim 7 wherein said foreground candidate mass filtering is performed using a Gaussian Laplacian Pyramid image decomposition method.
9. A method according to claim 7 wherein said foreground candidate mass filtering is performed using wavelet image decomposition.
10. A method according to claim 7 comprising segmentation by iso-level contouring of the filtered result and by applying rules to the resulting contour maps.
11. A method according to claim 1 wherein said foreground candidate mass filtering comprises a blob detection process at multiple scales.
12. A method according to claim 11 whereby segmentation of a mass is obtained as the central portion of the blob having a highest convolved filter response among said scales.
13. A method according to claim 11 , wherein blobs are local increases in image density, the blobs being modeled as a circular shaped intensity surface including a central maximum and decaying in all directions outwards from the center.
14. A method according to claim 1 wherein said foreground candidate mass filtering comprises at least one even symmetric Gabor filter or at least one bandpass Gabor filter.
15. A method according to claim 14 comprising segmentation by iso-level contouring of the filtered result and by applying rules to the resulting contour maps.
16. A method according to claim 14 using a single Gabor filter.
17. A computer executing a computer program product on a computer-readable medium adapted to detect candidate masses in a mammographic image, program product causes the computer to subject the mammographic signal representing said image to a cascaded filtering process, wherein the cascaded filtering process comprises: 1) background removal filtering for removing a globally varying component in said mammographic image, which is based on a physical model of breast composition and/or breast compression and includes a minimum tracking filter from thorax side to mammographic skin line side in which minimum filter values of the minimum tracking filter spread out in a radial way until the breast skin line is reached, and 2) foreground candidate mass filtering, and subject a result of said cascaded filtering to segmentation.
18. A computer software product for detection of candidate masses in a mammographic image, the product comprising a non-transient computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to subject the mammographic signal representing said image to a cascaded filtering process, wherein the cascaded filtering process comprises: 1) background removal filtering for removing a globally varying component in said mammographic image, which is based on a physical model of breast composition and/or breast compression and includes a minimum tracking filter from thorax side to mammographic skin line side in which minimum filter values of the minimum tracking filter spread out in a radial way until the breast skin line is reached, and 2) foreground candidate mass filtering, and subject a result of said cascaded filtering to segmentation.
19. A method of detecting candidate masses in a mammographic image, comprising:
subjecting the mammographic signal representing said image to a cascaded filtering process, wherein the cascaded filtering process comprises background removal filtering for removing a globally varying component in said mammographic image and foreground candidate mass filtering comprising a blob detection process at multiple scales, wherein the blob detection process includes a computation of a reduced resolution image from the mammographic image, computation of a normalized Laplacian, computation of maxima of Laplacian in an image domain, and computation of a maxima of a Laplacian in a scale-space, and
subjecting a result of said cascaded filtering to segmentation.
20. A method of detecting candidate masses in a mammographic image, comprising:
subjecting the mammographic signal representing said image to a cascaded filtering process, wherein the cascaded filtering process comprises background removal filtering for removing a globally varying component in said mammographic image, wherein said background removal filtering is based on a physical model of breast composition and/or breast compression and comprises a minimum tracking filter from thorax side to mammographic skin line side in which minimum filter values of the minimum tracking filter spread out in a radial way until the breast skin line is reached; and
subjecting a result of said cascaded filtering to segmentation.