Evaluation of breast density variability between right and left breasts

Breast density is an established risk factor for developing breast cancer. In clinical routine it is qualitatively assessed by visual inspection of mammographies by radiologists on the basis of the American College of Radiology (ACR) Breast Imaging Reporting and Data System (BI-RADS) guidelines. This might lead to intra- and inter-observer variability. As a consequence, many Computer-Aided Diagnosis (CAD) systems, based on machine learning and artificial intelligence, have been developed in recent years to support radiologists in breast density evaluation. One difficulty arises when considering the evaluation of the performance of a CAD. It is well known that very large data sets must be used to obtain a reliable assessment of generalization ability. However, according to ACR guidelines, if the breasts appear not to be of equal density only the denser category should be reported; therefore it is not possible to access to densities of both breasts when analyzing retrospective data. This leads to difficulties in gathering large data sets of clinical mammographic images. As a large number of clinical breast images data sets have been acquired using BI-RADS one can pose the question if it is possible to train an automatic system using one single label for both breasts potentially different. In this study, we evaluated the what extent differences in density between breasts of the same patients occur in a real population. In particular we extracted multiple radiomics features from breast parenchyma using the Pyradiomics library and a preliminary clustering analysis has been conducted. Preliminary results showed different densities between right and left breast occur in 25-30% of cases. © 2023 Convegno Nazionale di Bioingegneria. All rights reserved.