Comprehensive Summary
This study compares breast cancer detection performance for mammogram screening when screened by breast radiologists with and without AI-based computer-aided detection (AI-CAD). Between February 2021 and December 2022, regular mammography screening data was collected from 24,543 women ages 40 years or older. A total of 24,545 mammograms, including 2 cases of bilateral breast cancer, were analyzed in this study. Breast cancer was detected with and without AI-CAD and the performance was compared. The researchers found that breast radiologists (BRs) who used AI-CAD detected 140 cancers in total, which is 17 more than BRs without AI-CAD who detected 123 breast cancers in total. The PPV1 is defined as the percentage of all positive screening exams with a pathologic cancer diagnosis and the PPV1 of BRs with AI- CAD was higher (12.6) compared to BRs without AI-CAD (11.2). The cancer detection rate (CDR) for BRs with AI-CAD was significantly higher by 13.8% compared to BRs without AI- CAD. In addition, interpreting mammograms with AI-CAD led to detecting 6 additional cases of ductal cancer in situ (DCIS), 11 additional cases of invasive cancer, a significant increase in the detection of small-sized cancer less than 20 mm, node-negative metastasis, luminal A subtype, and lower grade invasive ductal carcinoma (IDC) when compared to the performance without AI-CAD. Overall, this study finds that BRs use of AI-CAD for screening mammograms results in significantly increased CDRs compared to BRs who don’t use AI-CAD.
Outcomes and Implications
While mammography screening has proven to be effective in detecting early cancer, breast cancer diagnosis can be delayed due to inaccurate false negative interpretation of mammography, failure to recognize the cancer, and misinterpretation or a wrong interpretation. To overcome these issues, AI can be used for mammography screening with the support of radiologists to improve the effectiveness of screening. AI-CAD can be used to enhance diagnostic performance when it comes to clinical breast cancer screening thus resulting in earlier detection for patients. Before this work can be applied in a clinic, a 2-year follow-up is needed to evaluate the true impact of AI-CAD use on interval cancers detected after two years and whether there is an increase in interval cancers with poor prognosis.