Research On Brain Tumor Segmentation Algorithm Based On Hybrid Clustering

Brain tumor is one of the most common major diseases in human beings,and the high incidence and lethality of brain tumor have caused people’s attention.Segmentation of brain lesions is essential for doctors’ accurate diagnosis and precise treatment.However,it exists some problems such as the complexity of the brain tissue structure,the noise and inconspicuous edges due to interference from medical imaging equipment.Therefore,how to accurately segment the lesion area is still difficult point and challenge.How to improve the accuracy of segmentation has become a hot topic.At present,a variety of medical image segmentation methods have been proposed at home and abroad,mainly including segmentation algorithms based on threshold,the segmentation methods based on deep learning,and the segmentation methods based on clustering.The clustering algorithms are very popular among the scholars because of their efficiency and unsupervised classification.Among the clustering algorithms,the fuzzy Cmeans(FCM)algorithm is the most common algorithm.The FCM algorithm introduces ambiguity,which is a kind of fuzzy flexible division.And this method is simple to implement.However,the traditional FCM algorithm is sensitive to noise,outliers and clustering parameters leading to instability,which is easy to lead to inaccurate segmentation results.This article focuses on the medical image segmentation algorithm which is sensitive to noise and outliers,as well as clustering instability.The brain tumor image segmentation algorithm is studied in depth,and the effect of accurate segmentation of the lesion area is achieved.The specific research content is as follows:(1)In order to improve the instability of traditional FCM algorithm and the susceptibility to noise interference,which lead to inaccurate segmentation,this paper proposes a hybrid clustering segmentation algorithm based on Gaussian kernel.Firstly,the adaptive Wiener filtering is adopted to remove the noise.Secondly,the K-means ++algorithm is used to initialize the clustering center,which can improve the stability of the cluster to some extent.Then,a Gaussian kernel function is introduced into the traditional FCM algorithm to reduce the sensitivity of the clustering process to parameters and further improve the stability of the algorithm.Finally,the brain lesion area is extracted based on thethreshold value.Experiments are performed on brain MRI images with noise of different intensities.The results show that the proposed segmentation algorithm is more stable and more accurate than the K-means algorithm,FCM algorithm,sFCM and csFCM algorithm.(2)In order to further improve the robustness of the clustering algorithm to noise,we introduce the fast-guided filter in the pre-processing and clustering process.First of all,a combination of adaptive Wiener filtering and fast guidance filters are used to denoise,which not only effectively removes noise but also retains detailed information of the original image under the guidance of the guided image.Secondly,SLIC superpixel pre-segmentation is adopted to avoid the existence of outliers,which effectively solves the problem that the clustering algorithm is sensitive to outliers.Then,a fast guided filter is introduced into the proposed hybrid clustering segmentation algorithm based on Gaussian kernel as a postprocessing for each iteration to further improve the robustness to noise.Finally,the tumor region is extracted based on the threshold value.Experiments are performed on images containing Gaussian noise of different intensities.The results show that compared with the improved algorithm in recent years,the algorithm can more effectively avoid the impact of noise and isolated points to the accuracy of brain lesions segmentation.