Osteoarthritis (OA) is the most common musculoskeletal disorder and presents a large societal burden. Knee pain in patients with knee OA is a leading contributor to physical disability and a major reason for hospital visits. An improved understanding of the etiology of knee pain has been hampered in part by knee OA being a multifactorial and progressive disease of the whole joint; consequently, knee pain progression may be the result of local or regional abnormalities of several different structural features over time. The long-term goal is to accelerate the development of optimal screening for enrollment into clinical trials to test promising treatments for symptom improvement. The overall objective in this application is to study the association of different MRI-based features with the temporal patterns of various knee pain measurements (e.g., knee pain frequency and severity) in OA. The central hypothesis is that there are some temporal knee pain phenotypes and various MRI-defined structural features (e.g., bone marrow lesions) are associated with the phenotypes. This hypothesis is formulated largely based on the preliminary studies, including the Osteoarthritis Initiative (OAI), the Multicenter Osteoarthritis Study (MOST), the semi-quantitative (SQ) readings, the complex knee pain measurements in the OAI and MOST studies, and projects on machine/deep learning to accurately predict SQ readings for MRIs that do not have existing radiologist-derived readings in the OAI and MOST studies. The central hypothesis will be tested by pursuing two specific aims: 1) identify different temporal knee pain phenotypes based on all available longitudinal knee pain measurements and the related knee pain risk factors in the MOST and OAI; and 2) associate the MRI-defined structural features at baseline with the identified temporal knee pain phenotypes. The research proposed in this application is innovative in several ways. It considers various definitions of knee pain and the available pain measurement data in the super- large longitudinal OAI and MOST studies and applies machine learning, deep learning and statistical methods to identify knee pain phenotypes and associate them with MRI-based factors. This new and substantively different approach to understanding knee pain is expected to overcome the limitations of existing studies (e.g., single knee pain measurement-based and cross-sectional studies), thereby opening new horizons for detecting different temporal knee pain phenotypes and allowing identification of individuals at high risk of various temporal knee pain phenotypes for more targeted enrollment into clinical trials.