Study On The Articular Cartilage Segmentation Algorithm Based On PSO-SVM

Early diagnosis of osteoarthritis is very important, it’s clinical manifestations areusually cartilage degeneration and disappearance.Through the articular cartilageimaging and image processing, we can capture the changes of cartilage and achieveearly diagnosis. Automatic segmentation of the articular cartilage images is animportant part of whole image processing, with advantages of objectivity, quantifiableetc. Unfortunately, at present there are very few published reports on automaticsegmentation method of articular cartilage images.Most of existing methods relate tosingle method,so their effectiveness are not satisfying, so it is very necessaryto researchfor an efficient automatic segmentation method for articular cartilage images.Based on the analysisabove, with the knee MRI sequences as the research object,we make use of a variety of image segmentation algorithm, to achieve an efficientmethod of automatic segmentation for articular cartilage images.This method is used tosegment the cartilage tissue accurately and quickly, so that it can provide a powerfultool for clinicians to calculate the thickness and volume of cartilage tissue，and observethe changes of the thickness and volume of different periods, for diagnosis andtreatment. There exist the problems of complex texture, serious noise, fuzzy boundariesbetween cartilage and non-cartilage in articular cartilage image, affecting thesegmentation results. In order to solve these problems, we did the main works asfollows:1) First, study and propose the improved adaptive Canny edge detection operator,by calculating the gradient value of each image anditeration to get the image’s bestthreshold; thenproposes a method of PSO-SVM to classify the edges in order to solvethe problem of missing and seized the edge extraction, thus better reserving the cartilageedge and achieving the precise positioning of cartilage.2) With the new similarity criteria based on difference of regional gray,solve theineffective problem of traditional region growing algorithm relying on the local natureof the image to determine the growth of fixed criteria. In addition,we use PSO tooptimize the adjustment coefficient of the newsimilarity criterion,improving theperformance of the improved region growing algorithm.Finally,we organizeexperimentsto compare this improved region growing method with some other segmentation method.The research results of this thesis will provide automatic segmentation of articularcartilage image a new solution，so as toprovide a new theoreticalbasis andmethods forpromotingthe early diagnosisof arthritis based on images, and has a certain theoreticalvalue and practical significance.