AI Can Be the Second Mammographic Reader

Background: Although screening mammogram guidelines have often recommended interpretation by multiple readers, limited availability of qualified readers has affected this capability. Artificial intelligence (AI) methods of mammographic images have shown promise in accurately interpreting these examinations. Objective: To determine whether AI as a second reader (AISR) can accurately interpret mammographic images as the second reader. Methods: Data from screening mammograms performed between 2005 and 2021 with ≥12 months’ follow-up from a Japanese cancer center were obtained. Mammograms were evaluated by 5 senior and 5 junior board-certified readers. Three methods to evaluate the mammograms compared with the AI assistance were used including the routine read, the AISR, and the AI-assisted junior reader. Results: The test set consisted of mammograms from 4059 women with 157 positive and 3902 negative. The area under the curve was 0.84 for the test set, not significantly differe more...