Research On Algorithms Based On Fuzzy Clustering Theory For Brain MR Image Segmentation

Abstract:Brain MR image segmentation is the basis of kinds of advanced brain image processing,and is the primary task of brain functional study.Brain MR image segmentation plays an important role on brain development and on the study and diagnosis of brain disease,so it can help doctor to formulate follow-up treatment and recovery plan.This paper firstly introduces the fuzzy set theory,and discusses the relationship between fuzzy set and fuzzy clustering.This paper introduces the Hard C Means(HCM) algorithm,and then the generated Fuzzy C Means Clustering algorithm which is used more widely.Based on the Lagrange Multiplier Method,this paper induces the objective function optimization problem to desmonstrate its convergence conditions. Several estimition parameters of segmentation quality are introduced,which are used to evaluate the performance of different algorithms.This paper introduces the concept and properties of kernel functions.Then the distance in the objective function of the FCM algorithm is improved with the use of Gaussian kernel,in order to strengthen the validity of the algorithm.This paper introduces a Probabilistic C Means(PCM) algorithm,this algorithm has no membership constraints,and different category has its corresponding probabilistic coefficient.This paper improves PCM algorithm with the use of kernel function,and then we get a Kenel Based Pobabilistic C Means(KPCM) Algorithm.This paper makes a discussion on the factors which affect the speed of fuzzy segmentation algorithm, and then puts forward a new initialization method to initialize the fast fuzzy clustering algorithm(HFFCM).This method uses the Gaussian function to filt the image histogram so to initialize the algorithm.Image histogram statistic method is used to reduce the caculation amount of the HFFCM algorithm.Experiment results show that the speed of HFFCM algorithm is much faster than several other basic algorithms,and its segmentation quality is more or less the same with FCM algorithm.This paper improves fuzzy clustering algorithms with spatial information,and then introduces the DFCM algorithm,sFCMpq algorithm.At last,a Fast Fuzzy Clustering Algorithm Based On Adaptive Median Theory(AMFFCM) is presented.This algorithm makes use of the principle of DFCM and sFCMpq algorithms,use adaptive median theory to improve the anti-noise performance of the algorithm,simultaneously the HFFCM algorithm is used to initialize this AMFFCM algorithm.The expriment results reveal that the segmentation quality of AMFFCM algorithm is better than other most algorithms,and the segmentation speed has been improved at the same time.