AI-driven Multimodal Knee Osteoarthritis Detection, Progression Prediction and Personalized Rehabilitation Recommendation System

Abstract

Knee osteoarthritis (KOA) is a chronic, progressive musculoskeletal disorder characterized by structural deterioration of the knee joint, pain, stiffness, reduced range of motion, muscle weakness, impaired physical function, and limitations in activities of daily living. The condition represents a major cause of disability and loss of mobility, particularly among older adults and individuals with obesity, previous joint injury, or prolonged mechanical loading. The knee is the most frequently affected joint in osteoarthritis, and a substantial proportion of individuals with osteoarthritis may benefit from rehabilitation interventions.Conventional KOA assessment generally relies on a combination of clinical examination, patient-reported symptoms, functional assessment, and medical imaging, particularly plain radiography. Radiographic severity is commonly characterized using the Kellgren–Lawrence (KL) grading system. Although radiographic assessment provides valuable information regarding structural joint changes, imaging findings alone may not fully represent an individual's pain, functional impairment, or future disease trajectory. Furthermore, patients with similar radiographic severity may experience substantially different symptoms and functional limitations.The heterogeneous nature of KOA creates a significant challenge for disease progression prediction and individualized treatment planning. Conventional statistical models may have limited ability to capture complex nonlinear relationships between imaging characteristics, demographic variables, clinical symptoms, functional status, behavioral factors, and longitudinal disease changes. Therefore, there is a need for intelligent computational approaches capable of integrating heterogeneous patient information within a unified predictive framework.