Research On Brain Magnetic Resonance Image Segmentation Algorithm Combining Guided Filter And Improved Fuzzy C-means

In recent years,brain diseases have gradually become a major hidden danger that affects the health of our nationals.According to statistics,more than one million people in our country die from brain diseases every year.With the development of medical imaging technology,magnetic resonance(MR)imaging has been widely used in the detection of brain function and brain diseases due to its high efficiency and safety.Brain MR images have become an important medium for in-depth research and analysis of brain tissue and brain diseases.Among them,the segmentation of brain MR images is used as the preprocessing link of subsequent brain image analysis,and its segmentation effect directly determines the results of follow-up research,which is a key step for brain tissue research and clinical diagnosis.Therefore,it is very necessary to propose an efficient brain MR image segmentation method.Based on the in-depth study of the existing brain MR image segmentation algorithms,this paper combines the characteristics of the brain MR image itself to study the fuzzy C-means(FCM)algorithm.The FCM algorithm is a soft clustering algorithm that can assign each pixel in an image to two or more categories.The ambiguity of the FCM algorithm makes it widely used in the segmentation of brain MR images.However,FCM algorithm also has the problems of susceptibility to noise interference,random generation of initial parameters,and lack of edge information in the segmentation results when segmenting brain MR images.Based on this,this paper proposes a brain MR image segmentation algorithm combining guided filter and improved FCM.First,the texture features of the image is combined with the gray features to form the fusion features.Then comprehensively consider the neighborhood density and distance relationship of pixels to adaptively select the initial cluster centers.After that,the fusion features are used as the feature constraint of the algorithm clustering iteration.Finally,a guided filter algorithm is used to correct the clustered pixels to make up for the loss of edge detail information.This paper includes the following two innovations:(1)Aiming at the problems of noise sensitivity and the randomness of initial clustering centers in the FCM-based brain MR image segmentation algorithm,an improved FCM brain MR image segmentation algorithm based on Tamura texture feature is proposed.The algorithm first combines the Tamura texture features and grayscale features of the image with linear weighting to form fusion features to increase the sharpness of the image,and then uses the fuzzy neighborhood density of pixels and the distance relationship between pixels to adaptively select the initial cluster centers,and finally uses the fusion features as the feature constraint of the similarity calculation between pixels to segment the image.This algorithm improves the algorithm’s anti-noise and segmentation accuracy,while reducing the number of iterations,effectively reducing the time consumption of the algorithm.(2)Aiming at the problem of the lack of edge detail information in the segmentation of brain MR images by fuzzy clustering algorithm,a pixel correction method based on guided filter algorithm is proposed.This method uses the original image as the pilot information for the guided filter algorithm,splits the optimal membership degree obtained from the segmented image,and performs edge enhancement and pixel correction processing on the split membership matrix,which further improves the algorithm’s segmentation accuracy.In the experiment,the algorithm proposed in this paper was used to segment two data sets of simulated brain MR image and real brain MR image,and its performance was analyzed from the three dimensions of algorithm anti-noise,segmentation accuracy and operating efficiency.The experimental results show that the segmentation accuracy of this algorithm can reach 90.73%in simulated brain MR images with high levels of noise and gray inhomogeneity;in real brain MR images,the segmentation accuracy of this algorithm can reach 92%.The algorithm proposed in this paper can obtain the initial clustering centers closer to the clustering results,can exhibit better robustness when processing noise and gray inhomogeneity images,and can achieve higher image segmentation accuracy while reducing the time-consuming operation of the algorithm.In summary,the algorithm proposed in this paper can quickly and effectively achieve accurate segmentation of brain MR images.