Evaluation of Automated Mammographic Density Classification in Tomosynthesis: Comparison with Radiologists

Abstract

Objective: Breast cancer screening is a valuable field of health research conducted through mammography. However, mammography evaluation is the examination with the most frequent lack of to agrement among radiologists. In this study we aimed to show the compatibility of mammographic density classification with a new software, Bellus Breast Density Measurement Software (Option), with visual examination. Methods: The mammographic density classification of 500 patients was retrospectively determined by five radiologists with varying levels of experience, according to the 5th version of the breast imaging reporting and data system (BIRADS). The mean age of 500 women included in the study was calculated as 53.8±10.08. The obtained data were compared with the Bellus software mammographic density classification of the same patients. Then, the visual evaluation and the compatibility of the Bellus software and the readers were compared. Results: The agreement between the Bellus software and all five readers was poor (kappa value 0.07-0.12). The agreement of the readers with each other is moderate-good (kappa value 0.054-0.64). The Intraclass Correlation Coefficient (ICC) value for the five separate readers was calculated to be 0.80, indicating good compatibility, while the ICC value for the Bellus software with the five separate readers was calculated to be 0.74, indicating moderate compatibility. The Friedman test revealed that while the mammographic density classification of each reader remained consistent, the classification provided by the Bellus software differed. Conclusion: Bellus Breast Density Measurement Software (Option) diagnostic accuracy is lower than visual examination. We recommend that the manufacturer develop the software.