ON THE ASSOCIATION BETWEEN DEEP LEARNING ASSESSED MRI MORPHOLOGICAL PHENOTYPE AND 4-YEARS CHANGES IN CARTILAGE THICKNESS AND T2 RELAXATION TIMES

Abstract

Purpose: Previous studies have leveraged deep learning to identify morphological phenotypes as well as to compute T2 relaxation times and change in cartilage thickness on the Osteoarthritis Initiative (OAI) knee MRI dataset. Deep learning-generated ROAMES phenotyping of osteoarthritis (OA) has been demonstrated to predict incidence of structural and symptomatic OA as well as likelihood of total knee replacement (TKR) within 96 months. While associations vary by cartilage compartment, T2 relaxation times have also been associated with OA incidence and TKR after 5 years.