Artificial Intelligence for Rheumatoid Arthritis and Osteoporosis Diagnosis from Radiography, Computed Tomography, and Multimodal Clinical Data: A Systematic Review

Abstract

Artificial intelligence (AI) is increasingly used to support musculoskeletal diagnosis from routine imaging, yet evidence remains fragmented across diseases, modalities, tasks, and validation designs. This systematic review synthesizes AI-based diagnosis, screening, grading, scoring, and structural assessment of rheumatoid arthritis (RA) and osteoporosis/osteopenia using radiography, computed tomography (CT), ultrasound, DXA-related imaging, and multimodal clinical data. Following PRISMA 2020 principles, Scopus, PubMed, Web of Science Core Collection, IEEE Xplore, and the ACM Digital Library were searched. Across the final exports, 2,239 records were identified; 247 full-text reports were assessed for eligibility, and 233 studies were included, while 25 reports could not be retrieved. CNN-based and transfer-learning approaches remained dominant, with increasing use of attention mechanisms, transformers, ensembles, explainable AI, and multimodal fusion. Osteoporosis research focused mainly on opportunistic screening and BMD-related prediction, whereas RA studies more often addressed diagnosis, joint-level damage, severity grading, and automated scoring. Methodological maturity remained uneven: 141 of 233 studies (60.5%) relied on internal validation only, whereas 44 (18.9%) reported external or multicenter validation. Future progress therefore depends less on marginal gains in headline metrics than on reproducible study design, robust external validation, calibration, and clinically integrated multimodal decision support.