Reducing false positives in computer-aided detection
View Patent ↗Methods, systems, and related computer program products for computer-aided detection (CAD) of anatomical abnormalities in digital (or digitized) x-ray mammograms are described. The inventive techniques are based on using a foundational CAD processing algorithm that is characterized by at least one of non-shift-invariance, non-rotational-invariance, and non-inversional-invariance. According to one preferred embodiment, a first x-ray mammogram image of a breast is received, and at least one altered version thereof is generated that differs therefrom by at least one of image shift, image rotation, and image inversion. The first x-ray mammogram image and each of the at least one altered versions thereof are individually processed using the foundational CAD algorithm to generate a respective plurality of individual CAD detection sets. The plurality of CAD detection sets are then compared to generate an overall CAD detection set.
1. A method for computer-aided detection (CAD) of anatomical abnormalities in x-ray mammograms based on a foundational CAD processing algorithm, the foundational CAD processing algorithm being characterized by at least one of non-shift-invariance, non-rotational-invariance, and non-inversional-invariance, the method comprising:
receiving a first x-ray mammogram image of a breast in digital form;
processing by digital computer the first x-ray mammogram image to generate at least one altered version thereof that differs therefrom by at least one of image shift, image rotation, and image inversion;
individually processing the first x-ray mammogram image and each of the at least one altered versions using the foundational CAD algorithm to generate a respective plurality of individual CAD detection sets; and
comparing the plurality of CAD detection sets to generate an overall CAD detection set for the first x-ray mammogram image,
wherein the foundational CAD algorithm is non-deformation-invariant in at least one direction.
2. The method of claim 1 , wherein comparing comprises, for each location in the breast identified for CAD marking in at least one of the CAD detection sets, including that breast location in the overall CAD detection set based on a collective evaluation of outcomes for that breast location in all of the CAD detection sets.
3. The method of claim 2 , wherein the collective evaluation of outcomes for the breast location comprises a voting technique in which the breast location is included in the overall CAD detection set only when at least seventy percent (70%) of the CAD detection sets have identified the breast location for CAD marking, and wherein the method further comprises displaying the overall CAD detection set to a viewer in conjunction with the first x-ray mammogram image on an output display.
4. The method of claim 1 , wherein the at least one altered version includes at least ten (10) successively rotated versions of the first x-ray mammogram image having rotation angles distributed between about −5 degrees and 5 degrees.
5. The method of claim 1 , the first x-ray mammogram image having a number of pixels in a range of about 1K×1K to about 10K×10K pixels, wherein the at least one altered version includes at least ten (10) successively shifted versions of the first x-ray mammogram image having shift amounts distributed between about 2 pixels and 20 pixels.
6. The method of claim 1 , the foundational CAD algorithm exhibits non-shift invariance by being at least partially non-robust against shifts of about 1 to about 20 pixels in the first x-ray mammogram image.
7. The method of claim 1 , the foundational CAD algorithm exhibiting non-rotational invariance by being at least partially non-robust against rotations of about −5 degrees to about +5 degrees in the first x-ray mammogram image.
8. The method of claim 1 , wherein the first x-ray mammogram is processed to generate a single altered version of the first x-ray mammogram that differs from the first x-ray mammogram image by at least one of image shift, image rotation and image inversion.
9. The method of claim 8 , the single altered version being a flipped version relative to the first x-ray mammogram image.
10. The method of claim 1 , wherein the foundational CAD algorithm is non-robust against small stretches or shrinkages along at least one direction.
11. A method for computer-aided detection (CAD) of anatomical abnormalities in x-ray mammograms based on a foundational CAD processing algorithm, the foundational CAD processing algorithm being characterized by at least one of non-shift-invariance, non-rotational-invariance, and non-inversional-invariance, the method comprising:
receiving a first x-ray mammogram image of a breast in digital form;
processing by digital computer the first x-ray mammogram image to generate at least one altered version thereof that differs therefrom by at least one of image shift, image rotation, and image inversion;
individually processing the first x-ray mammogram image and each of the at least one altered versions using the foundational CAD algorithm to generate a respective plurality of individual CAD detection sets; and
comparing the plurality of CAD detection sets to generate an overall CAD detection set for the first x-ray mammogram image,
the foundational CAD algorithm exhibiting non-inversional invariance by being at least partially non-robust against mirroring or flipping spatial inversions of the first x-ray mammogram image.
12. A method for computer-aided detection (CAD) of anatomical abnormalities in x-ray mammograms based on a foundational CAD processing algorithm, the foundational CAD processing algorithm being characterized by non-shift-invariance, nonrotational-invariance, and non-inversional-invariance, the method comprising:
receiving a first x-ray mammogram image of a breast in digital form;
processing by digital computer the first x-ray mammogram image to generate a plurality of altered versions thereof including at least one spatially inverted version thereof, at least one rotated version thereof, and at least one shifted version thereof;
individually processing the first x-ray mammogram image and each of the plurality of altered versions using the foundational CAD algorithm to generate a respective plurality of individual CAD detection sets; and
comparing the plurality of CAD detection sets to generate an overall CAD detection set for the first x-ray mammogram image.
13. The method of claim 12 , wherein comparing comprises, for each location in the breast identified for CAD marking in at least one of the CAD detection sets, including that breast location in the overall CAD detection set based on a collective evaluation of outcomes for that breast location in all of the CAD detection sets.
14. The method of claim 13 , wherein the collective evaluation of outcomes for the breast location comprises a voting technique in which the breast location is included in the overall CAD detection set only when at least seventy percent (70%) of the CAD detection sets have identified the breast location for CAD marking, and wherein the method further comprises displaying the overall CAD detection set to a viewer in conjunction with the first x-ray mammogram image on an output display.
15. The method of claim 12 , wherein the at least one rotated version includes at least ten (10) successively rotated versions of the first x-ray mammogram image having rotation angles distributed between about −5 degrees and 5 degrees.
16. The method of claim 12 , the first x-ray mammogram image having a number of pixels in a range of about 1K×1K to about 10K×10K pixels, wherein the at least one shifted version includes at least ten (10) successively shifted versions of the first x-ray mammogram image having shift amounts distributed between about 2 pixels and 20 pixels.