Unlocking the potential of training: enhancing inter-reader agreement in background parenchymal enhancement assessment on contrast-enhanced mammography

Abstract Objective This study aimed to evaluate the inter-reader agreement of visual background parenchymal enhancement (BPE) assessment on contrast-enhanced mammography (CEM) and determine whether training can improve this assessment. Methods Five hundred and forty-eight women who underwent contrast-enhanced mammography from 2018 through 2022 were included. A total of 2135 images were read by six radiologists (Group A) independently and rated BPE with the ordinal scale: minimal, mild, moderate, or marked. After a dedicated training, the same cases were reread in a new random order within two weeks. To further verify the effectiveness of training, four radiologists (Group B) from four different institutions were invited to assess BPE in the same way with 800 images. The accuracy of each reader and agreement on BPE categorization between radiologists and the reference standard were evaluated. Results The percentage of correct BPE assignments increased from 62.3% (range, 48.6%–73.6%) to 75.5% (range, 69.0%–79.7%) after training in Group A and increased from 55.7% (range, 52.0%–58.9%) to 66.9% (range, 62.0%–71.0%) after training in Group B. With training, inter-reader agreement increased from fair to moderate in both Group A (κ = 0.32–0.52) and Group B (κ = 0.40–0.53), indicating that readers learned and the training can be rolled out to different institutions. Conclusion The initial inter-reader agreement in assessing BPE among radiologists was fair but showed enhancement following training, highlighting the necessity of training and standard sets of reference images that should be made available for training. Advances in knowledge This study demonstrates that dedicated training improves inter-reader agreement in BPE assessment on CEM, highlighting the need for standardized training programs to enhance consistency in clinical practice and research.