Investigating the performance of a CAD<inf>x</inf> scheme for mammography in specific BIRADS categories

Abstract

A Computer Aided Diagnosis (CAD x ) pipeline has already been introduced to discriminate between benign and malignant clusters of microcalcifications (MCs). In this study, we evaluate the specific methodologies using cases from publicly available databases of mammograms, the MIAS database and the Digital Database of Screening Mammography (DDSM). Specifically, we investigate various subsets of regions of interest (ROIs) containing cluster of MCs, following the BIRADS assessment performed by radiologists who have participated in the preparation of the databases. The obtained results indicate that there are specific BIRADS categories where the proposed system provides high classification results. Furthermore, it seems that the best results are achieved in obscure cases, where the radiologists have doubts about their diagnosis and recommend extra medical examinations and short follow-up. The reported results indicate the potential of the proposed CAD x system to assist the diagnostic task of radiologists by providing a reliable second opinion concerning the diagnosis of a cluster of MCs.

Fused man-machine classification schemes to enhance diagnosis of breast microcalcifications

“…In a previous study [19], we indicated that CAD x methodologies perform satisfactorily in subsets of obscure cases (such as BIRADS 3), where the radiologists recommend a short follow-up. This observation implies that such 'oriented' schemes, following radiologist's BIRADS assessment, may be adopted during the diagnostic process to assist expert's diagnosis.…”

A Statistical Approach to Assess the Robustness of Radiomics Features in the Discrimination of Mammographic Lesions

“…Studies from the literature have shown that a double reading from two independent radiologists can improve the diagnosis of breast lesions [18][19][20][21]; however, a double check has a high operational cost in terms of resource utilization and time, thus preventing its application in broader healthcare settings. In this sense, CAD systems, along with reproducible and standardized radiomics workflows, could offer the opportunity to obtain a "second opinion" from an automated tool, thus improving the diagnostic accuracy with less operational cost and within a reduced timeframe [22,23]. The sensitivity of such automated tools, i.e., the ability to find and properly characterize suspicious areas, is, therefore, of fundamental importance to increase the reliability and clinical translation of CAD systems and radiomics workflows in most radiology applications [24][25][26][27][28][29][30][31][32], which can be then combined with artificial intelligence (AI) approaches, such as machine learning (ML) and deep learning (DL) techniques, to further boost the automated diagnostic performance [33][34][35][36][37][38][39][40][41].…”