Local-based Fuzzy Clustering Algorithm For Segmentation Of MR Brain Images

Image segmentation is often required as a preliminary and indispensable stage in thecomputer aided medical image process. Unfortunately, intensity inhomogeneities in MRimages, which can change the local statistical characteristics of the image, and cause thebrightness distribution of different physiological organizations overlap, are a major obstacle toany automatic methods for MR image.In this paper, a local-based fuzzy clustering algorithm (LBFC) for Magnetic Resonancebrain images corrupted by intensity inhomogeneity was proposed. This method is based onthe fuzzy clustering algorithm, the global threshold segmentation algorithm, the theory ofShannon entropy and the distributing disciplinarian of brain tissues in anatomy.A local image model is defined to eliminate the adverse impact of both artificial andinherent intensity inhomogeneities in magnetic resonance imaging on intensity-based imagesegmentation methods. The estimation and correction procedures for intensityinhomogeneities are no longer indispensable because the highly convoluted spatialdistribution of different tissues in the brain is taken into consideration.On the basis of the local image model, we propose a tissue-based local entropy minimummethod to generate the contexts for all pixels. Firstly, a local-based global thresholdsegmentation algorithm is presented to obtain the white matter image and cerebrospinal fluidimage of the processing MRI. And based on the distributing disciplinarian in anatomy thatgray matter (GM) is always between white matter (WM) and cerebrospinal fluid (CSF) inbrain, the two images can be regarded as the reference to judge whether the two matters existtogether in a clustering context, and to confirm that all the three tissues exist together. Thenthe Shannon entropy is used as a homogeneity measurement to optimize the size of theclustering context. Finally, according to the suitable of fuzzy c-means (FCM) algorithm forthe uncertainty and fuzziness of gray-scale image, FCM algorithm is independently performedin each context to calculate the degree of membership of a pixel to each tissue class. Becausethe centroids in two neighboring contexts are approximately equivalent, so the computationaltime can be reduced by using the cluster centroids of the previous context as the initial valuesof the current one.The efficacy of the proposed algorithm is demonstrated by extensive segmentationexperiments using both simulated and real MR images and by comparison with otherpublished algorithm.