AI-DRIVEN AND MULTIMODAL INNOVATIONS IN BIOMEDICAL IMAGING AND SENSING IN ARTHRITIS

Abstract

Artificial intelligence (AI) especially deep learning is changing biomedical imaging and quantitative sensing for all types of arthritis such as osteoarthritis (OA), rheumatoid arthritis (RA), psoriatic arthritis and axial spondyloarthritis (axSpA). Recent advances include automated grading of structural damage on X-rays, three-dimensional (3D) segmentation and tissue characterization on MRI and CT, radiomics-based phenotyping and combining imaging with clinical, biomarker and multi-omics data to predict prognosis and treatment response. Deep learning models now match or exceed the performance of expert readers in detecting cartilage lesions, automating Kellgren–Lawrence grading, segmenting cartilage and menisci and predicting radiographic progression and the need for joint replacement from baseline MRI or radiographs in osteoarthritis. AI helps with automated radiographic scoring, MRI detection of inflammatory sacroiliitis and risk stratification for cardiovascular complications in RA and spondyloarthritis using imaging-derived radiomic biomarkers. AI pipelines are being used more and more to analyze multimodal workflows in musculoskeletal imaging that combine radiography, CT, DXA, MRI, ultrasound and clinical data. These pipelines help find quantitative, reproducible biomarkers of joint damage, inflammation and prognosis. This narrative review talks about new AI-driven and multimodal technologies for imaging and sensing arthritis. It also talks about new clinical uses that are coming up and the main problems that need to be solved such as generalizability, bias, interpretability and regulatory translation.