ARTIFICIAL INTELLIGENCE IN MUSCULOSKELETAL RADIOLOGY: FROM IMAGING BIOMARKERS AND FOUNDATION MODELS TO PRECISION DIAGNOSIS AND PERSONALIZED CARE

Abstract

Musculoskeletal radiology is undergoing a paradigm shift with the incorporation of artificial intelligence (AI) that moves imaging from morphology-driven interpretation to a quantitative, biologically oriented and individualized assessment. This review aims to explore the evolving role of AI in musculoskeletal imaging, highlighting how imaging biomarkers, molecular and genomic data, multi-omics data, foundation models, and precision medicine converge. Automated detection, segmentation, classification, and extraction of quantitative phenotypes from bone, cartilage, muscle, and soft tissues using AI analysis of radiography, computed tomography, magnetic resonance imaging, and ultrasound. Combination of these imaging phenotypes with genomics, transcriptomics, proteomics, metabolomics, and clinical information offers a chance to describe disease heterogeneity and link the imaging with the molecular mechanisms involved. Radio genomic and multimodal strategies can help identify biologically different disease sub types, composite radio imaging–molecular biomarkers, individualized risk profiles and predictors of the disease progression or therapeutic response. Foundation models also enable further possibilities, such as transferable representations and cross-modal learning that can incorporate a wide range of biomedical data. Important applications of these developments are in the field of osteoarthritis, osteoporosis, inflammatory musculoskeletal diseases, tumors, and muscle diseases. But it is hindered by data heterogeneity, scarcity of paired imaging–omics data, reproducibility, interpretability, bias, privacy, and external validation. Imaging, genomic, and molecular data coupled with standardized longitudinal clinical data will be crucial for the future of AI-driven precision diagnosis and personalized musculoskeletal care in future multicentre studies.