Performance of Artificial Intelligence Tools in Axial Spondyloarthritis Imaging Assessment: a Systematic Literature Review and Meta-analysis.
Authors
Affiliations (5)
Affiliations (5)
- DESiRE-consulting, Sorée, Belgium.
- Dept. of Rheumatology, Ghent University hospital, Ghent, Belgium.
- Dept. of Radiology, Ghent University hospital, Ghent, Belgium; Dept. of Diagnostic Sciences, Faculty of Medicine and Health Sciences, Ghent University, Ghent, Belgium.
- Dept. of Rheumatology, Ghent University hospital, Ghent, Belgium; Center for Inflammation Research, Molecular Immunology and Inflammation Unit, VIB-Ghent University, Zwijnaarde, Belgium.
- DESiRE-consulting, Sorée, Belgium. Electronic address: [email protected].
Abstract
Advances in Artificial Intelligence (AI) have opened new opportunities for improving detection, as well as the accuracy and efficiency of imaging interpretation in axial spondyloarthritis (axSpA). The aim of this study is to summarize the performance of AI techniques versus human reader in interpreting imaging modalities (magnetic resonance imaging (MRI), computed tomography (CT) and conventional radiography (CR)) in axSpA. In line with the PRISMA guidelines, a systematic literature review was conducted across PubMed and Scopus for studies published between 1 January 2010 and 7 June 2025. Individual performance metrics were extracted and analyzed using a meta-analysis approach. For the meta-analyses, the overall estimate was computed using the DerSimonian-Laird random effect model on both subject and image levels. Heterogeneity was assessed using Higgings I². A total of 1033 references were identified, 46 full texts were reviewed, and 33 studies were included. All studies were published between 2020 and 2025, with 58% in 2024/2025. Sixty-seven % of the studies originated from Asia. Most of the studies included MRI (64%) and applied Deep learning techniques (85%). Overall performance estimates (95%CI) at subject level were: sensitivity 87% (85; 90%), specificity 80% (75; 85%), accuracy 84% (81; 87%) and receiver operating characteristic area-under-the curve 0.88 (0.86; 0.90). On image level the results were similar. The number of studies assessing AI performance in axSpA using imaging modalities has increased tremendously in the past years. Although AI showed good performance, human expert remains essential to reach diagnostic accuracy while maintaining clinical safety.