Method to increase sensitivity and specificity of computer-aided detection system for mammographic images in dense breast parenchyma

Abstract

Aim. To search ways to increase the diagnostic value of computer-aided detection of pathological lesions for mammography based on the principle of comparing the images of contralateral breasts.Methods. Analysis of the diagnostic value of computer-aided detection of pathological lesions for mammography MammCheck 1.15 of our own design, which included asymmetric regions and the brightness transformation search algorithms, was performed. To test this system standard digital mammograms in craniocaudal and mediolateral oblique views of 117 patients with morphologically verified breast cancer (visualized as focal lesions with or without microcalcifications) and 114 patients who did not have malignant tumors, which was confirmed by the results of a 3-year follow-up, were used. All mammograms had a density 3-4 (C-D) according to the ACR classification. In 23 of the 117 patients, visualized changes corresponded to breast cancer were blur or generally not visible with unaided eye on standard mammograms.Results.Method overall sensitivity was 80.3%, false positive rate - 13.2%. Sensitivity in identifying lesions with microcalcifications was higher (100%) compared to the lesions without microcalcifications (78.1%, pConclusion. The breast parenchyma density remains a problem for the computer-aided detection of pathological lesions in cancer diagnosis, especially when not accompanied by the microcalcifications, however, these systems can detect malignant lesions, which are invisible or barely visible in the standard study, and therefore their use is advisable as an option for a second or third mammograms reading.