Knee osteoarthritis (OA) affects 10% of the population over 55 years old, and is a major cause of work absence, early retirement, and joint replacement. The purpose of this study is to explore the possibility of using a recently proposed knee osteoarthritis biomarker and machine learning method to predict OA progression. The biomarker, named cartilage damage index (CDI), was extracted from 3D knee MR images by measuring representative locations. In total, the CDI measured 60 locations on the whole knee cartilage, including 18 from femur, 18 from tibia, and 24 from patella. The CDI values at these 60 locations in the baseline year were used as feature input to the artificial neural network (ANN). We trained two types of model to predict OA and OA severity change, respectively. The label of each sample was “OA” or “non-OA”, based on the severity level in two years when training the model for OA prediction; and the label of each sample was “change” or “no-change” based on the severity level change in two years, when training the model for OA change prediction. Separate ANN models were trained for three OA severity measures, i.e., lateral Joint Space Narrow (JSN), medial JSN and Kellgren and Lawrence (KL) grade. Besides using all the 60 informative locations, we tested different combinations of CDI locations at sub-regions. The best prediction result achieved in this work was AUC (area under ROC curve) = 0.912, using the whole knee CDI (all 60 locations) to predict lateral JSN. Experiment results showed that CDI can be used as a reliable cartilage quantification method and help to predict OA progression in two years.