Project Abstract Osteoarthritis (OA) is the most common musculoskeletal disorder, presenting a large societal burden and affecting over 300 million worldwide. Knee pain in patients with knee OA is a leading contributor to physical disability and a major reason for hospital visits. Prior studies that have examined the determinants of knee pain have not considered the complexity of knee pain reporting systematically (e.g., localized, regional or diffuse knee pain). Moreover, the existing approaches focus on accurate prediction of knee pain and have limitation on the interpretability. There is, therefore, a critical need to address this crucial void by developing the computational tools with accurate prediction and interpretable properties. The overall goal of this research is to is to build computational tools to accurately predict different incident frequent knee pain spatial patterns and interpret the association between these knee pain patterns and the structural abnormalities (i.e., Bone Marrow Lesions (BML)) detected on Magnetic Resonance Imaging (MRI). The central hypothesis is that early in the disease course, abnormalities of structural features on knee MRI will be determinants of specific knee pain spatial patterns. This study is innovative in that it focuses on incident knee pain, examining knee pain spatial patterns early in the disease course, and taking into account different definitions of knee pain. This study is also innovative in that the study team will leverage excellent prediction performance of deep learning and interpretable properties of statistical approaches to predict and understand different knee pain spatial patterns. The specific aims are 1) To adapt a Transformer-based deep learning approach for predicting the incident localized, regional, or diffuse knee pain from MRIs in the Osteoarthritis Initiative (OAI), and 2) to develop a novel interpretable statistical framework for characterizing the relationship between BMLs on MRIs and incident localized, regional, or diffuse knee pain.