We conducted a literature search using the Embase (Ovid), MEDLINE (Ovid), and Web of Science Core Collection databases. Search terms addressed the concepts of AI, machine learning, and cancers of the breast. We used Covidence to screen and organize search results. The inclusion criteria consisted of randomized controlled trials (RCTs) in which AI was used in the screening, prognosis, diagnosis and/or treatment of breast cancer, and/or in support of patients with breast cancer. Six reviewers screened titles and abstracts for each manuscript, with two reviewers independently screening each study, and we resolved conflicts via group discussion. Four reviewers read the full text for the remaining studies, with two independent reviewers per manuscript, resolving disagreements through discussion. From each study, we extracted information about the intervention used, outcomes assessed, and results published. We conducted a Cochrane risk-of-bias assessment for each study.

Our search yielded 5,882 studies, of which six RCTs met our inclusion criteria. We excluded most papers due to being retrospective studies (3,920), with others being off topic/not written in English (1,049), reviews/meta-analyses (408), editorials (131), case reports (22), surveys (21), datasets (24), or protocols (26). Among 196 prospective studies, six were RCTs. For these six RCTs, three utilized AI to aid in diagnosis using either mammography or ultrasound, while the rest involved medical counseling, psychological support, and supportive care. Three studies reported significant results in their primary and/or secondary outcomes: (1) use of AI improved recall rate of mammogram readings; (2) AI improved the sensitivity, specificity, and accuracy of detecting lymph node metastasis in ultrasound readings; and (3) an AI app enhanced patients’ mental health and quality of life. We found two studies had high risk-of-bias, one had some concerns for bias, and three had low risk-of-bias.

We conducted a systematic review of the literature to describe the current utility of AI in breast cancer, and our findings highlight opportunities for future research. Specifically, we found that only six RCTs have been published to date investigating the use of AI in breast cancer. Our results demonstrate that AI has the potential to help with the diagnosis of breast cancer through mammography and ultrasound, while also helping support patients’ psychological health through mobile apps.