The Segmentation of Brain Magnetic Resonance Image by Adaptive Fuzzy Support Vector Machine

Accurate segmentation of brain magnetic resonance (MR) image is very important for brain disease diagnosis. Unfortunately, due to image characteristics as intensity non-uniformities and fuzzy edges, it is hard to achieve good results by using traditional brain magnetic resonance (MR) image segmentation methods. In this paper, an adaptive fuzzy support vector machine (AFSVM) is constructed for brain MR image segmentation. AFSVM introduces adaptive parameters related to the attenuation rate and critical value of fuzzy membership, thus defining a new membership function to alleviate the problem that the traditional FSVM is ineffective to depict the distribution characteristics of brain MR images. Experimental test shows that, for brain MR images with different noise level and weighted imaging modalities, the classification accuracy rates of the proposed method are higher than those of BP neural network, the classical SVM or the traditional FSVM. Moreover, the computational complexity of the proposed method is also acceptable.