Synthesizing evidence regarding artificial intelligence generated radiological reports based on medical images: a scoping review protocol.

IntroductionConsidering numerous radiological images and the heavy workload of writing corresponding reports in clinical work, it is significant to leverage artificial intelligence (AI) to facilitate this process and reduce the burden of radiologists. In the past few years, particularly with the advent of vision language models, some works explored generating radiological reports directly from images. However, despite some efforts demonstrated in previous studies, limitations in AI-generated radiological reports persist. Current research mainly focuses on detecting abnormalities, rather than generating textual reports from medical images. The evidence for AI application in radiological report writing has not been synthesized. This scoping review aims to map the current literature on the engagement of AI-generating radiological reports based on images. Methods and analysisFollowing a well-established scoping review methodology, five stages are provided: i) determining the research question, ii) searching strategy, iii) inclusion/exclusion criteria, iv) data extraction, and v) results analysis. Four databases will be applied to search peer-reviewed literature from January 2016 to February 2025. A two-stage screening process will be conducted by two independent reviewers to determine the eligibility of articles, and only those regarding AI-generated radiological reports will be included. All data from eligible articles will be extracted and analyzed using narrative and descriptive analyses, presenting in a standard form. Ethic and disseminationEthic approval is no required in this scoping review. Experts from Hospital of University of Geneva will be consulted to provide professional insight and feedback regarding the study findings and help with dissemination activities in peer-reviewed publications or academic presentations