Abstract
In this study, we extended our previous work on analyzing the relationship between cartilage thickness and osteoarthritis (OA) severity grade change. Cartilage thickness is measured by the Cartilage Damage Index (CDI) which includes 60 points marked on 3D MRI for each knee joint. In our previous work, we used CDI points on femur and tibia compartments only (36 points) as features and employed machine learning methods to predict the OA severity grade change. In this work, we added the 24 CDI points on patella into the feature space and explored whether CDI points from patella could improve the accuracy, on a larger dataset. Kellgren-Lawrence (KL) grade was used to measure OA severity in this study. Artificial neural network (ANN), which showed good performance in our previous study, was employed as the machine learning method. For KL grade classification, experiment results showed that adding patella points improved the performance remarkably, from AUC 0.822 to AUC 0.903 and the whole knee CDI achieved the best classification performance on the dataset.