Osteoarthritis (OA) is a chronic and debilitating joint disease characterized by degenerative changes to the bones, cartilage, meniscus, ligaments and synovial tissue. Since the OA is one of the most common disease among adults, costs of treatment have large economic impact for the patients and also the health care system. One of the main effects of OA is degradation of articular cartilage, which leads to pain and mobility loss of the joints. Articular cartilage of the knee is a thin layer that covers the end of bones in the joint. In OA the cartilage is damaged. Therefore, one of the effective methods to assess disease progression is monitoring of cartilage changes such as shape, volume or thickness. According to the reports acquired from radiologist and orthopedics during diagnosis of articular cartilage defect based on MRI imaging and the orthopedics observation some mismatches appear between them. What causes this phenomenon is the intrinsic ambiguity of cartilage MRI images in showing the edges and features of the cartilage precisely. However based on image processing of MRI images the disease progression can be diagnosed and programmed before knee surgery. This procedure is performed by bone and cartilage segmentation of knee MRI. In this thesis, several methods are proposed for automatic segmentation of bones in knee MRI’s. The segmentation methods are based on active shape models and active contours. Initialization of the contour is obtained from image registration techniques. The methods are tested on 3 slices of 20 datasets from SKI10 database and the results have been compared with medical diagnosis done by experts. The method based on active shape models obtained an average Dice similarity coefficient (DSC) of 0.9382, 0.9077 for femur and tibia bones and 0.9601, 0.9458 obtained with active contour. Keywords: Osteoarthritis, segmentation, image registration, active shape model, active contour
