AI More Accurate Than Human Readers in DBT Interpretation

Background: Breast cancer screening programs using digital breast tomosynthesis (DBT) improve cancer detection and reduce recall rates over digital mammography (DM); however, DBT interpretation time is almost twice that required for interpreting DM and strategies to improve efficiency and accuracy are needed. Computer-aided detection (CAD) has been used as an adjunct to mammography interpretation, but traditional CAD with user-designed descriptors led to high false-positive rates and decreased use. More recent artificial intelligence (AI)-based CAD systems have showed better performance. There is insufficient evidence on the independent performance of AI for DBT to support its implementation into clinical workflows. Objective: To develop and evaluate an AI model for the diagnosis of breast cancer on DBT images, and to determine whether it could improve accuracy and reduce reading times. Methods: During a 12-year period, a deep learning AI algorithm was developed and validated for DBT more...