The Research Of Medical Image Segmentation Based On Fuzzy C-Means Algorithm

Medical image segmentation is the preparation technology to image preprocessingand pattern recognition,such as feature registration,feature quantification,3Dreconstruction and etc. The effect of segmentation will impact the results of computerassisted diagnosis. In this thesis, the brain magnetic resonance (MR)imagesegmentation is researched, the research on the paper focuses on the problem ofclassing the normal brain tissues.Fuzzy C-means (FCM) clustering algorithm is a classical one in fuzzy clusteranalysis algorithms, using iterative optimization of the objective function to obtain thefuzzy partition data sets, and is good at convergence. FCM algorithm can avoid settingthe threshold, solve a number of branches of the partition problem, and suit imageswith ambiguity and uncertainty. As an unsupervised clustering algorithm FCMalgorithm does not need manual intervention and is very suitable for the automatic ofimage segmentation. But Standard FCM algorithm is taking into account the grayinformation, not the spatial information of the image, so the algorithm is sensitive tonoise.Considering that medical image data must contain much noise in the process ofacquisition, so the designed algorithm must be robust to noise.In this paper, focusing on improved FCM clustering algorithm and its applicationin image segmentation, the following work is done:Firstly, this thesis explores the medical image segmentation of standard FCMalgorithm and studies such issues as the selection of the initial number of clustering,and the identification of the initial center and initial membership matrix. We alsomainly present many improved FCM methods in recent years in the lecture. They aregenerally classed into three kinds: the first one, the constraints on membership functionis changed, the second one, the term of spatial information is introduced, the third one,the Kernel method is introduced, and the typical ones of these algorithms are analysedand appraised simply.Secondly, because the standard FCM algorithm does not consider the space pixelsinformation and is sensitive to noise, an improved algorithm for image segmentationbased on objective function is proposed. The improved algorithm is good use of theneighborhood pixels by introducing an adaptive weighted coefficient which is used tocontrol the effect of the neighborhood pixels on central pixel, and realized by modifying the objective function given in the Chen’s algorithm.To realize fastclustering,the beginning of the algorithm use Fast Fuzzy c-means.Experiment resultsshow that the improved algorithm is more efficient and more robust to noise than thestandard FCM and FCM_S1.Finally, because the standard FCM algorithm does not consider the space pixelsinformation and is sensitive to noise, an improved algorithm for image segmentationbased on membership function is proposed. Two kinds of spatial information areincorporated in the membership function of the standard FCM, A priori probabilityand Fuzzy spatial information. A priori probability is automatically determined in theimplementation of the algorithm by the fuzzy memberships. Fuzzy spatial informationis the average of fuzzy membership of the neighborhood pixels to a cluster. To realizefast clustering, the beginning of the algorithm use Fast Fuzzy c-means. We proposedthe algorithm, which incorporate spatial information into FCM, have shownconsiderable resilience to noise, yet with increased noise levels in images, theseapproaches have performed exceptionally well.