Commercially available AI medical devices for prostate MRI: current evidence and opportunities for implementation

Abstract Objective Artificial intelligence (AI) may improve prostate MRI interpretation and reduce radiologist workload, but evidence supporting clinical implementation remains limited. This scoping review summarises evidence on commercially available AI medical devices for prostate MRI. Materials and methods PubMed was searched for studies published through October 1, 2026. Eligible devices were cleared by the United States Food and Drug Administration and/or CE marked, with commercial availability corroborated through manufacturer websites. Data extraction addressed standalone AI performance, AI-assisted radiologist performance, workflow efficiency, and downstream clinical impact. Results Twenty-one studies evaluating 7 commercial AI systems were included; 17/21 (81%) were retrospective and 14/21 (67%) were single-centre. AI assistance suggested greater benefit for less-experienced readers, but improvements were inconsistent and increased specificity sometimes accompanied reduced sensitivity. Standalone evaluations, outside the devices’ approved intended use, showed discrimination comparable to that of radiologists in selected cohorts. Inter-reader agreement and workflow benefits were mixed. We identify three key limitations of the current body of evidence that require thoughtful consideration before routine implementation: (1) incomplete verification strategies, (2) limited determination of workflow benefit, and (3) lack of analyses of cost-effectiveness and patient outcomes, e.g., time to treat or cancer-free survival. Conclusion Evidence supporting the use of commercial AI for prostate MRI remains limited and heterogeneous, with inconsistent diagnostic benefits. Prospective comparative studies in real clinical workflows are needed to determine its potential role within the prostate cancer diagnostic pathway. Key Points Question Evidence for clinical implementation of prostate AI remains limited—most studies are retrospective and single-centre, with heterogeneous populations and reference standards. Findings AI may improve diagnostic performance for less-experienced readers, but gains for expert readers and assessment of workflow efficiency are limited. Clinical relevance Prospective multicentre studies embedded in real-world pathways are needed, together with analyses of cost-effectiveness and patient outcomes .