Translational efficiency of artificial intelligence in breast cancer: A systematic review of research output and clinical trial translation (2015–2025)

Artificial intelligence (AI) research in breast cancer has expanded rapidly; however, its translation into prospective clinical evaluation remains unclear. This systematic review evaluated the translational efficiency (TE) of AI applications by comparing research publications with interventional clinical trials across domains. PubMed/MEDLINE and ClinicalTrials.gov were searched for studies published between January 2015 and December 2025. Studies applying machine learning or deep learning to breast cancer detection, diagnosis, prognosis, or treatment decision-making were included. Eligible studies were categorized by application domain, and TE was calculated as the ratio of interventional trials to research publications. Domain distributions were compared using Fisher’s exact test. Out of 293 publications identified, 180 met inclusion criteria, along with 36 clinical trials. Imaging AI predominated in research output (90/180, 50.0%), followed by clinical decision support (19/180, 10.6%) and risk/prognostic AI (19/180, 10.6%). Among clinical trials, screening (14/36, 38.9%) and digital health/patient support (12/36, 33.3%) were most represented. TE varied significantly across domains, with the highest values in digital health/patient support (109.1%) and screening (82.4%), moderate translation in clinical decision support (36.8%), and low or absent translation in imaging (2.2%), pathology (6.3%), and risk/prognostic AI (0%) ( P < 0.001). AI translation in breast cancer remains uneven, with implementation concentrated in screening and digital health despite a predominance of imaging-based research, highlighting a critical translational gap and the need for clinically driven validation strategies.