Advancing Knee Osteoarthritis Assessment Using Novel Quantitative Imaging Features and Machine Learning

Abstract

Knee osteoarthritis (KOA) is a leading cause of disability worldwide, characterized by progressive degeneration of articular cartilage and associated musculoskeletal structures. Despite its prevalence, early diagnosis and patient-specific assessment remain challenging due to the multifactorial nature of the disease and the limitations of conventional imaging and grading systems. This thesis advances KOA evaluation by integrating multi-modal quantitative imaging features and machine learning methodologies to improve diagnostic accuracy, reproducibility, and early detection. Magnetic resonance imaging (MRI) and computed tomography (CT) scans were acquired and analyzed within the framework of the EU SINPAIN project, which aims to develop innovative and personalized therapies for KOA. The imaging data were processed through a standardized pipeline for segmentation, registration, and feature extraction of bones, cartilages, and muscles. Morphological, density, and radiomic features were combined to train and validate supervised and semi-supervised learning models and identify the most informative imaging biomarkers of KOA. The study demonstrates that specific imaging-derived characteristics, such as alterations in cartilage morphology and texture, subchondral bone density distribution, and the balance between muscle and intramuscular adipose tissue, serve as sensitive indicators of degenerative changes in the joint. Cross-tissue analyses revealed consistent interdependence between cartilage degradation and surrounding musculoskeletal adaptations, supporting a holistic interpretation of joint health. Furthermore, robustness and stability assessments established the reliability of the extracted quantitative features under realistic segmentation variations. By uniting advanced imaging analytics with artificial intelligence, this work contributes a reproducible framework for objective and personalized assessment of KOA. The developed methodology enhances the understanding of tissue-level interactions in joint degeneration and supports the transition toward precision diagnostics and targeted therapeutic strategies in osteoarthritis care.