Research on the segmentation of tiny multi-target in brain tissues based on support vector machines

The support vector machine (SVM) algorithm is applied to segment caudatum, putamen and pallidum region in brain magnetic resonance imaging (MRI) in this paper. A multi-classification SVM based on two-classification SVM is proposed in the segmentation processing. Firstly, the rough sets (RS) and principal component analysis (PCA) are separately used for reducing the dimension number of the high dimensional feature vectors extracted from Brain MRI. Secondly, the multi-classification SVM are adopted to classify for the non-reduction high dimensional feature vectors and the reduced feature vectors respectively. Finally, the classification performance of the multi-classification SVM is analyzed according to the false alarm probability, the false dismissal probability and the segmentation accuracy. A great deal of experimental results shows that the segmentation accuracy of the proposed multi-classification SVM segmentation is the highest compared with the k-means clustering (KMC), the fuzzy c-mean clustering segmentation (FCMS), k-nearest neighbor method (KNN), the Bayes classifier and the radial basis function neural network (RBFNN) segmentation for any feature vectors. However, the high segmentation accuracy is gotten at the cost of high computational complexity.