Breast tissue density estimation using automated volumetric breast density compared to radiologists’ visual assessment by Breast Imaging Reporting And Data System (BIRADS) at Charlotte Maxeke Johannesburg Academic Hospital (CMJAH), South Africa

Abstract

There are different methods of assessing breast tissue density, which include automated volumetric assessment and visual assessment using the BI-RADS classification. The aim of the study was to assess concordance in evaluation of breast density between fully automated volumetric and visual assessment by radiologists. Secondarily, intra-observer concordance was assessed for each reader, re-reading 100 of the images after 8-weeks. A retrospective study using images from 2017 and 2018 was conducted at Charlotte Maxeke Johannesburg Academic Hospital. The sample included 192 women. The mammograms were performed on the Full Field Digital Mammogram Hologic Selenia Dimensions. The mammograms were reviewed by four radiologists using BARCO 5 and BARCO 3 megapixel monitors. The volumetric density was obtained using Hologic Quantra 3 software. Agreement assessed with kappa statistics. Experienced readers agreed with the software in 46.9% and 45.8% of the cases (kappa 0,174 and 0.132). The less experienced readers showed 40% and 36.4% (kappa 0.075 and 0.074). Intra-reader concordance was 35% and 60%, (kappa 0.016 and 0.365). The second inter reader agreement observed between the radiologists’ second reading and the machine was assessed. Readers 1 and 2, 2nd reading showed an agreement of 47% and 49% (kappa 0.1519 and 0.1638). Readers 3 and 4, 2nd reading was 41.00% and 37% (kappa 0.1146 and 0.0601). The study showed that visual assessment of breast density on mammograms is observer depended. Experienced radiologists demonstrated slightly higher agreement with automated assessment than junior radiologists. There was also intra-reader variation when readers were asked to assess the same cases 8-weeks later