Research On Brain MR Image Segmentation Based On FCM Algorithm

The latest statistics of National Health and Family Planning Commission showed that brain diseases such as cerebral hemorrhage,brain tumors,cerebral infarction and brain trauma have become the first killer threatening to people's life and health.Therefore,the prevention and treatment for brain disease is one of the key focuses in the field of medical research.As a non-invasive medical diagnostic method to obtain the information of the human body about internal tissues,medical imaging technology can not only alleviate the suffering of patients,but also bring great convenience for the doctors and researchers.As one of medical imaging technologies,magnetic resonance imaging(MRI)has the advantages of no ionizing radiation and can be used for the detection of almost all human organs,so it has been widely used,especially the MR images of brain and spinal cord can provide more practical value.In the course of clinical diagnosis of the doctors,in order to detect the physiological or pathological state of brain tissue volume changes,it is very important to accurately segment brain tissues in the brain MR image.Generally,brain MR images are divided into three parts: white matter,gray matter and cerebrospinal fluid.The doctors can quantitative measure the cross-sectional area or volume of the brain tissue or lesion,and improve their diagnosis accuracy for the patient's symptoms,and provide a reference for the further diagnosis of the treatment plan.In recent years,with the rapid development of medical imaging technology,massive image data is placed in front of the doctor,early manual segmentation method has been unable to adapt to the current needs,and the automatic segmentation of medical image using computer-assisted method has become more mainstream and acceptable.However,the brain MR images are often affected by some factors such as the performance of the imaging equipment,the nonuniformity of the RF field strength and the changes of the patient's position during the imaging process.There are often noise,intensity inhomogeneity,and low contrast of brain tissue in MR images,which brings interference to precise brain tissue segmentation.Therefore,this paper focuses on these problems,several effective segmentation schemes of brain MR image based on fuzzy c-means clustering algorithm were designed and developed,and the main work is as follows:(1)A fast robust fuzzy c-means clustering algorithm is proposed during segmentation process of brain MR images.According to the gray information of the pixels in the local neighborhood,the neighboring pixel deviations are calculated in the kernel space based on the grayscale medium value,and the normalized adaptive weighting measure of each pixel is obtained.The pulse noise and Gaussian noise in the image can be effectively suppressed,and the details and edge information of brain MR images are better protected.At the same time,the single pixel in the image are replaced by gray histogram in the process of clustering,so the algorithm improves the segmentation accuracy while increasing the efficiency of fuzzy c-means clustering.(2)An unsupervised fuzzy c-means clustering algorithm based on nonlinear weighted local information of brain MR image is proposed.Considering the factors of spatial measure between pixels,the spatial distance and the gray level information between the local neighborhood pixels are combined in the non-linear weighting form in the similarity measure of the fuzzy clustering,the constrained relationship between the center pixel and its neighboring pixels can be more accurately described in the local area,and the objective function is more reasonably established.In addition,it is helpful to improve the clustering performance of the image by introducing the local neighborhood information into the fuzzy membership degree,and experiments have shown that the proposed algorithm effectively improves the segmentation accuracy and overcomes the problem of image degradation caused by the severe noise and outliers in brain MR images.(3)A fuzzy c-means clustering algorithm combining local and non-local information of brain MR image is proposed.In normal conditions,a set of samples with similar neighborhood configurations can be found for each pixel in the image,that is,pixel has non-local information with redundant features.In the Gaussian kernel space,the non-local information as a constraint term is adaptively combined with the local information,thus new objective function of fuzzy clustering is constructed,at the same time,bias field information of brain MR image is also coupled into the model.The model not only has strong anti-noise ability,but also can effectively eliminate the influence of bias field,and the brain tissue in MR image with intensity inhomogeneity can be accurately extracted.(4)A fuzzy c-means clustering algorithm incorporating Markov random field model is proposed to focus on noise,intensity inhomogeneity and low contrast in brain MR images.Because Markov random field can be used to describe the spatial correlation of the image,the local spatial information is combined into the distance measure,and a constraint term with neighborhood information is constructed.At the same time,a non-local regularization term with global feature is designed.Then,both terms are organically combined to establish the objective function of fuzzy clustering.This algorithm combines the randomness and fuzziness of the image segmentation problem,the prior knowledge of the image can be obtained reasonably.To an extent,it eliminates the influence of the low contrast in the MR image.In the meantime,the algorithm can estimate the bias field in the image and effectively improve the image segmentation accuracy.