The inter-reader agreement in the interpretation of mammography images according to BI-RADS by Moscow radiologists

Background: Breast malignancies take a leading position among incident cancers in women. Mammography has been recognized as the main method for early detection of breast cancer. However, mammogram assessments are based on a subjective opinion of the radiologist, which could lead to diagnostic disagreement. According to the literature, inter-radiologist agreement on mammograms varies from 0.450 to 0.888. Aim: To assess the inter-reader agreement in mammogram interpretation with BI-RADS (Breast Imaging Reporting and Data System) by radiologists of the Moscow city (Russia). Methods: The study included 741 mammography images done from January 15, 2020, to June 25, 2023. All mammograms were downloaded from the Unified Radiology Information Service of the Unified Medical Information and Analytical System (EMIAS) of the Moscow city and included radiologist reports with a BI-RADS score (the initial assessment). Each mammogram was further analyzed by two radiologists (with their job experience from 2 to 5 years) (this was the first revision) and thereafter by two more radiologists (with their job experience above 5 years and scientific degree) as a part of the expert review. The inter-reader agreement was assessed using an intra-class correlation coefficient. Results: The inter-reader agreement for the full BI-RADS score between radiologists who performed the initial assessment and those performing the first revision ranged from 0.836 [95% confidence interval (CI) 0.801–0.865] to 0.875 [95% CI 0.848–0.897]. Similar agreement was observed between radiologists who performed the initial assessment and the experts: 0.838 [95% CI 0.804–0.866] to 0.879 [95% CI 0.854–0.901]. The agreement on the full BI-RADS scale between radiologists who performed the first revision and the experts was significantly higher (p 0.001) than with those performing the initial assessment and ranged from 0.890 [95% CI 0.866–0.910] to 0.963 [95% CI 0.954–0.970]. Conclusion: The inter-reader agreement between radiologists of the Moscow city in the assessment of mammography study results on the full BI-RADS scale is high. The agreement between the radiologists who performed the revision is higher than their agreement with the radiologists who performed the initial assessment, which may indicate better and more stable results obtained during the revision.