Novel Approaches For Segmentation Of MRI Brain Images

A novel method for segmentation of brain tissues in MRI (Magnetic Resonance Imaging)images is proposed in this paper. Firstly, we de-noise the images using a versatilewavelet-based filter. Subsequently, watershed algorithm is applied to brain tissues as an initialsegmenting method. Normally, result of classical watershed algorithm on gray-scale texturedimages such as tissue images is over-segmentation. The following procedure is a mergingprocess for the over-segmentation regions using fuzzy clustering algorithm (Fuzzy C-Means).But there are still some regions which are not divided completely, particularly in thetransitional regions of gray matter and white matter, or cerebrospinal fluid and gray matter.We proposed a rule-based re-segmentation processing approach to partition these regionsbased on the combination of local region continuity and global information. This integratedapproach yields a robust and precise segmentation. The efficacy of the proposed algorithm isvalidated using extensive experiments.In this paper, we also proposed another new approach to segment the brain MRI imagesusing a multi-scale and adaptive spatial fuzzy self-organizing feature map.Kohonen's self-organizing feature map (SOM) is a two-layer feedforward competitivelearning network, and has been used as a competitive learning clustering algorithm in brainMRI images segmentation. However, most brain MRI images always present overlappinggray-scale intensities for different tissues, particularly in the transitional regions of graymatter and white matter, or cerebrospinal fluid and gray matter. Therefore, fuzzy methods areintegrated with SOM in this paper to overcome this problem. Moreover, for image data, thereis strong correlation between neighboring pixels. To produce meaningful segmentation, weproposed a multi-scale and adaptive spatial fuzzy self-organizing feature map (MSFSOM) forMRI image segmentation, in which we consider the spatial relationships between imagepixels and multi-scale processing method to reduce the noise effect and the classificationambiguity. The efficacy of our approach is validated by extensive experiments using bothsimulated and real MRI images.