3D MRI brain image segmentation: A two-stage framework

Abstract

Image Segmentation is a process of delineating an image into some meaningful regions. It has the significant impact on the computer guided medical image diagnosis and research. The Magnetic Resonance Imaging (MRI) brain data are severely affected by the noise and inhomogeneity artifacts which lead to blurry edges in the intersection of the intra-organ soft tissue regions. This paper presents a novel two stage framework for segmenting the 3D brain MR image data. The first stage consists of modified fuzzy c-means algorithm (MoFCM) which incorporates the spatial neighborhood information of the volume data to define the new local membership function along with the traditional fuzzy c-means (FCM) membership function. The cluster prototypes obtained from the first stage are fed into the modified spatial fuzzy c-means (MSFCM) algorithm which includes 3D spatial information of the 3D brain MR image volume to generate the final prototypes. Our main endeavor is to address the shortcomings of the traditional FCM which is highly sensitive to noise as it solely depends on the intensity values of the image and develop a new method which performs well in noisy environment. The method is validated on several simulated and in-vivo 3D brain MR image volumes. The empirical results show the supremacy of our method than the other FCM based algorithms devised in the past.