Artificial Intelligence for Breast Cancer Screening in Mammography (AI-STREAM): Preliminary Interim Analysis of a Prospective Multicenter Cohort Study

Abstract

While retrospective studies have shown that artificial intelligence (AI) improve mammography screening accuracy, prospective data, particularly in a single-read setting, is lacking. This study aimed to address this knowledge gap by assessing the diagnostic accuracy of radiologists, with and without an AI-based computer-aided detection algorithm (AI-CAD), for interpretating screening mammograms in a single-read setting. A prospective multicenter cohort study in six academic hospitals participant in Korea’s national breast screening program was done, where women aged ³40 years were eligible for enrolment between February 2021, and December 2022. Radiologists interpreting screening mammograms first without, followed by with AI-CAD, and compared cancer detection rates (CDRs) and recall rate (RRs) for breast radiologists, general radiologists, and standalone AI. Of 24,543 women aged ³40 years were included in the final cohort (mean age 61 years [IQR 51-68]), with 131 (0.53%) screen-detected cancers confirmed based on pathologic diagnosis within six months. The CDR was significantly higher by 13.7% for breast radiologists with AI-CAD (n=124 [5.05 ‰]) versus those without AI (n=109 [4.44 ‰]; p <0.001), with no significant difference in RRs (p =0.564). Similar trends were observed for general radiologist, with significant higher CDRs by 25.1% for those with AI-CAD (n=105 [4·28 ‰]) versus those without AI-CAD (n=84 [3·42 ‰]; p <0·001); the CDR of standalone AI (n=118 [4·81 ‰]) was also significantly higher than that of general radiologists without AI, with no significant differences in RRs (p =0·795). Findings from this prospective, multicenter cohort study demonstrated significant improvement in CDRs and unaffected RRs of breast radiologist when using AI-CAD, as compared to not using AI-CAD, when interpreting screening mammograms in a single-read setting, highlighting the positive effects of AI-CAD as an assistive diagnostic tool to help radiologists, regardless of experience, in a real-world, breast cancer screening population.