Study On The Automatic Segmentation And Quantitative Analysis Of The Cartilages From MRI Of The Knee

Osteoarthritis(OA) has become the most common joint disease affecting the quality of life among middle-aged and older adult. And the age of OA patient is gradually decreasing. The key of the treatment of OA is the early detection and diagnosis. With the development of technology of medical imaging, MRI has become the main method to diagnose the OA due to its refined tissue resolving power, non-invasion and non-radioactive. The segmentation and quantitative analysis of the cartilages from MRI of the knee can assist doctors to diagnose the patholofical stage of the cartilage, which has a great application prospects. In knee MRI, the anatomy of cartilages is various, small and narrow, and the intensities of cartilages and surrounding tissues are similar. In addition, MR artifacts, noise, obscure also increase the difficulty of cartilage automatic segmentation and quantitative analysis.For the multiple existing problems discussed above, this thesis proposed an automatic segmentation algorithm based on multiple features support vector machine(MFSVM) and elastic automatic region growing(EARG), and quantitative analysis with the automaticsegmentation results. The main contents are as follows:(1) Due to the difficulties of cartilage segmentation in knee MRI, this thesis studied and proposed a bone-cartilage interface extraction algorithm based on improved adaptive Canny edge detection and multi-feature support vector machine to localize the candidate regions of cartilages. The improved adaptive Canny edge detection detects the major edges in knee MRI by the two different thresholds that are adaptively calculated by the feedback edges and the gradient magnitude histogram. The bone edges are recognized by using support vector machine with multi-feature. The accurate bone-cartilage interfaces are obtained by using optimization algorithm.(2) The automatic cartilage segmentation is achieved by using elastic automatic region growing based on bone-cartilage interfaces. The initial seed points are automatically set based on bone-cartilage interfaces and gradient magnitude. The similarity criterion is updated by the mean and standard deviation of the region that already grown.The accurate cartilage segmentations are obtained by optimizing the primary segmentation results using the anatomic knowledge and morphology.(3) The three-dimensional reconstruction and quantitative analysis are carried out based on the automatic cartilage results. The comparison between the volunteer with no obvious signs of OA, the volunteer with mild signs of OA and the volunteer with severe signs of OA shows the differences of morphology with different pathological state. The quantitative analysis of volume and thickness can verify the effectiveness of the proposed segmentation and quantitative analysis algorithm.The proposed method provided a new idea, and a new theoretical basis and method basis for diagnosis of OA based on knee MRI. It has certain theoretical significance and application value.