Economic evaluations of AI applications in radiology: a systematic review.

objectivesArtificial intelligence (AI) applications in radiology may improve clinical outcomes, but adoption is hindered by limited health economic evidence. This review synthesises economic evaluations of radiology AI, mapping methods, outcomes and metrics to identify trends, support future research, and inform policy. MATERIALS AND

methodsA systematic search of MEDLINE and Cochrane Central (2014-2025) was conducted on 21st March 2025, following PRISMA guidelines and recommendations for economic reviews. Eligible studies were peer-reviewed full-text economic evaluations of radiology AI compared with standard of care or non-AI interventions. Exclusion criteria included invasive imaging techniques, applications not using AI and the cost of training AI models. Data were extracted into economic, patient, and clinical domains by three reviewers. Reporting quality was assessed using CHEERS-AI for decision-analytic models.

resultsThirty-one studies met the inclusion criteria, including sixteen full economic evaluations. Reported outcomes varied, most often focusing on direct costs, cost-effectiveness, and diagnostic accuracy. Quality-adjusted life years (QALYs) were the predominant measure, though alternatives such as cost per patient screened or cost per correct diagnosis were also used. Approximately half of the studies employed decision-analytic modelling, mainly in opportunistic imaging. Geographic distribution was skewed, with most originating from the US and UK, and limited evidence from continental Europe. Few studies assessed productivity, workflow efficiency, or access to care.

conclusionsStudies varied in design, comparators, and outcome measures: only 16 of 31 conducted full economic evaluations, and CHEERS-AI scores ranged widely (34-89). Few studies included productivity or workflow outcomes, highlighting areas for future research beyond diagnostic accuracy and direct costs. Harmonised international guidance and interdisciplinary collaboration are needed to generate robust, comparable evidence to support responsible AI adoption in radiology. KEY POINTS: Question What economic outcomes and metrics are currently captured in the evaluation of radiology AI? Findings Current economic evaluations of radiology AI show inconsistent outcome measures, highlighting the need for harmonised assessment standards. This review identifies common metrics to improve comparability, strengthen research, and guide responsible adoption. Clinical relevance Radiology AI studies should measure economic outcomes relevant for decision-making. Broader, standardised approaches, supported by international guidance and multidisciplinary collaboration, are essential to demonstrate value and enable safe, evidence-based integration into healthcare systems.