Osteoarthritis (OA) is the most common form of arthritis. Affected individuals commonly suffer with chronic pain, joint dysfunction, and reduced quality of life. OA also confers an immense burden on health services and economies. Current OA therapies are symptomatic and there are no therapies that modify structural progression. The lack of validated, responsive and reliable biomarkers represents a major barrier to the development of structure-modifying therapies. MRI provides tremendous insight into OA structural disease and has highlighted the importance of subchondral bone in OA. The hypothesis underlying this thesis is that novel quantitative imaging biomarkers of subchondral bone will provide valid measures for OA clinical trials. The Osteoarthritis Initiative (OAI) provided a large natural history database of knee OA to enable testing of the validity of these novel biomarkers. A systematic literature review identified independent associations between subchondral bone features with structural progression, pain and total knee replacement in peripheral joint OA. However very few papers examined the association of 3D bone shape with these patient-centred outcomes. A cross-sectional analysis of the OAI established a significant association between 3D bone area and conventional radiographic OA severity scores, establishing construct validity of 3D bone shape. A nested case-control analysis within the OAI determined that 3D bone shape was associated with the outcome of future total knee replacement, establishing predictive validity for 3D bone shape. A regression analysis within the OAI identified that 3D bone shape was associated with current knee symptoms but not incident symptoms, establishing evidence of concurrent but not predictive validity for new symptoms. In summary, 3D bone shape is an important biomarker of OA which has construct and predictive validity in knee OA. This thesis, along with parallel work on reliability and responsiveness provides evidence supporting its suitability for use in clinical trials.