Artificial intelligence and machine learning applications in Behçet’s disease: a scoping review protocol

Abstract

Abstract Background Behçet's disease (BD) is a chronic multisystem vasculitis of unknown aetiology with a characteristic geographic distribution along the historic Silk Road, with highest prevalence in Turkey, Iran, and surrounding regions. Diagnosis remains clinical and often delayed, and management requires individualised treatment decisions across heterogeneous organ involvement patterns. Artificial intelligence (AI) and machine learning (ML) applications have expanded rapidly in rheumatology and clinical medicine, and a growing number of studies now apply these methods to BD-related tasks including diagnosis, differential diagnosis, severity stratification, outcome prediction, and biomarker discovery. However, this literature is heterogeneous across data modalities, clinical tasks, ML methods, and validation strategies, and to our knowledge no dedicated review has yet mapped the current state of AI applications in BD. This scoping review will systematically map this evidence base to identify current applications, methodological approaches, reporting quality, and research gaps. Methods This scoping review will follow the methodological framework of Arksey and O'Malley, refined by Levac and colleagues, and the Joanna Briggs Institute guidance for scoping reviews. Reporting will follow PRISMA-ScR. Eligibility will be defined using the Population–Concept–Context (PCC) framework: Population—patients diagnosed with BD according to any recognised criteria; Concept—application of any AI or ML method (including classical machine learning, deep learning, natural language processing, and large language models); Context—any clinical or research setting addressing a BD-related clinical or translational question. PubMed/MEDLINE, Scopus, Web of Science Core Collection, IEEE Xplore, and ACM Digital Library will be searched from inception to the search execution date. Title/abstract and full-text screening will be performed by two independent reviewers using Rayyan, with conflict resolution by consensus and a third reviewer where needed. Records not published in English will be machine-translated at the title/abstract stage and screened on content rather than excluded on language alone; those judged potentially eligible will be listed in a supplementary appendix and excluded at the full-text stage with the reason recorded in the flow diagram. Data will be extracted using a structured form, with a seeded random 20% sample independently re-extracted against prespecified agreement thresholds. Quality of predictive-model studies will be appraised using PROBAST + AI, and imaging studies additionally against CLAIM, with results presented as domain-level traffic-light and summary bar plots. Synthesis will be descriptive: performance metrics will not be pooled, averaged, or ranked across studies, but reported as ranges within defined task–modality strata alongside validation design, with emphasis on validation and reporting practice. Results will be presented through structured tables, visual gap maps, temporal trends, and a narrative summary. Discussion To our knowledge, this will be the first dedicated scoping review of AI and ML applications in Behçet's disease. Expected outputs include a structured map of AI applications across clinical tasks and data modalities, identification of reporting and methodological gaps, geographic distribution of cohorts, and a research agenda for future original work. On the basis of the appraisal, we will propose a candidate minimum reporting and validation set that we would argue should be met before an AI tool is recommended for BD care in a rheumatology guideline such as a future update of the EULAR recommendations. Findings will inform both clinicians considering AI-assisted BD tools and researchers designing future AI studies in BD. The review is restricted to English-language publications, which may under-represent the high-prevalence Silk Road populations in which BD is most common; this limitation is examined in relation to algorithmic bias and model generalisation. The protocol was registered on OSF prior to any database search. The review will be executed with two independent reviewers, and findings will be disseminated through peer-reviewed publication. Reporting guideline Protocol prepared in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA-P) 2015 checklist (Additional file 1). The review itself will follow the PRISMA extension for Scoping Reviews (PRISMA-ScR) 2018 guidance. Systematic review registration Open Science Framework 10.17605/OSF.IO/4ZJYW