Research On Brain Medical Image Segmentation Based On Fuzzy Clustering Theory

Medical image segmentation has been widely concerned as a means of clinical adjuvant medical treatment recently,and it plays an important role in practical application.The most common used medical images include Magnetic Resonance Imaging(MRI),CT images,and ultrasound images and so on.Magnetic resonance imaging technology has been widely used for its non-invasive and low-cost advantages,especially for soft tissues with high resolution.Human brain is complicated and changeable,in recent years,brain diseases such as brain tumors,brain injuries and cerebral vascular disease have always seriously threatened human health,among them,brain tumors are the most common disease except the cerebral vascular disease.The key to the treatment of brain tumors lie in the resection of the tumor,and the segmentation of the tumor plays an important role in the panning of resection.It becomes the urgent need of the clinician to accurately locate the relevant tissues in the brain,accurately identify the relevant lesion area.Therefore,the segmentation of brain medical image has important research value and significance.The fuzzy theory has certain inevitability and rationality in solving the problem of image.In fact,the image is fuzzy correlation in essence,such as the definition of the edge and boundary of the image is ambiguous.Fuzzy clustering has been widely used in medical image processing,the main contents of this thesis can be summarized as follows:For brain images with strong noise,we propose an improved fuzzy clustering algorithm based on spatial information.The fuzzy c-means algorithm is introduced into the kernel distance function to replace the traditional Euclidean distance,and adds the space constraints and membership degree of punishment into object function,and distributes corresponding local space to constraint two terms for each pixel,finally,the membership function is integrated with the spatial information of the image itself,and the result of the final robust segmentation is obtained.This thesis shows that the fuzzy clustering algorithm based on kernel has better segmentation results for the brain images with strong noise.For brain tumor images with blurred boundaries and complex structure,we propose a fuzzy spectral clustering algorithm based on super pixels.In this thesis,the SLIC segmentation algorithm is used to divide the image into super pixels with uniform size and shape,and extract the statistical features of these super pixel blocks based on gray histogram,and obtain the corresponding eigcnmatrix.Then,the fuzzy similarity matrix is calculated and the undirected graph is constructed;then the Laplace eigenvectors of the graph are clustered to obtain the segmentation results.The fuzzy similarity matrix is obtained by the fuzzy similarity measure based on the feature of the super pixel.We propose self-adaption parameter to control the similarity,and remeasure the similarity between feature points to obtain the fuzzy similarity matrix.This thesis shows that the segmentation algorithm based on super pixel is effective in the segmentation of brain tumor images.