Abstract: We present an algorithm that automatically and accurately segments the cartilages from Magnetic Resonance Images (MRI) of knees using multi-feature Support Vector Machine (SVM) and elastic region growing. First, adaptive canny edge detection is used to extract the main edges of the images. Second, the edge features are extracted and the edges are classified with SVM to complete the location of the cartilage edges. Then, based on the cartilage edges, the seeds and regions of the cartilage are chosen, and the preliminary segmentation of knee cartilages is achieved using the elastic region growing. Finally, the priori knowledge and morphology are used to improve the preliminary segmentation results. Experimental results show that the proposed method can accurately and rapidly segment the knee cartilages with high consistency with manual segmentation results. The proposed method obtains average DSC values of 0.8543, 0.8280 and 0.8703 for the femoral, tibial and patellar cartilages.

Key words: information processing technology, multi-feature support vector machine, elastic region growing, automatic segmentation of knee cartilages