Study Of Mri Brain Tumor Image Segmentation Algorithm Based On Spectral Clustering

MRI brain tumor image segmentation can help us get pathological brain tumor regions from MRI images and provide information for doctors to diagnose the disease.Therefore,the study of brain tumor image segmentation is of great significance.Spectral clustering is a clustering method based on graph.Compared with traditional clustering method,it has the advantages of simple principle,no requirement on the distribution shape of data space and convergence to the global solution and so on,and it has achieved good results in image segmentation.However,due to the evolution of spectral clustering algorithm from graph theory,there are still some problems to be solved in similarity matrix construction,large-scale data processing and other aspects.In this paper,the spectral clustering method will be used to segment MRI brain tumor images,and some problems existing in this method will be improved to achieve better segmentation effect.The research contents of this paper are as follows:(1)The internal pathology of brain tumors is complex,and the features such as gray scale,size and texture of the lesion are constantly changing in the MRI images,which makes the brain tumors have the characteristics of diversity and complexity,which also increases the difficulty of segmentation.Aiming at this characteristic,this paper proposes a multi-feature fusion superpixel spectral clustering algorithm for MRI brain tumor image segmentation.Firstly,the algorithm uses the SLIC algorithm for image pre-segmentation,then use superpixels instead of single pixel structure weighted undirected graph,and construct the similarity matrix by the fusion of multiple image features,and then get the Laplacian matrix.Lastly,we get the segmentation results by carrying out K-means clustering on Laplacian matrix.Experiments show that this algorithm can reduce the complexity of the algorithm and get more accurate segmentation of MRI brain tumor images.(2)Color image contains more information than gray image,and its storage structure and representation are more complex,so that the segmentation difficulty is further increased.Therefore,aiming at color MRI brain tumor images,this paper proposes a color MRI brain tumor image segmentation algorithm based on Nystrom extended spectrum clustering based on joint sampling.Firstly,the similarity matrix is constructed by using Nystrom sampling approximation strategy which combine four features of the superpixel,including color,space,covariance and edge based on the superpixel segmentation.At the same time,due to the expansion of the Nystrom algorithm,the selection of sampling points will affect stability and accuracy of the algorithm,so the interval sampling is adopted in this algorithm combined with artificial sampling way of sampling,it can not only obtain more representative sample point,also is helpful for reflecting the internal characteristics of image,obtain accurate segmentation results.The experiment shows that the proposed algorithm can get satisfactory segmentation results and has better stability.