Research On Image Segmentation Algorithm On The Brain Tumor Image Of MRI

Brain cancer is a high-risk disease.Surgical resection is an effective treatment in all therapeutic methods.Any information about location,size,and structure is crucial.Medical imaging techniques can be used to detect abnormal changes in body tissue.The magnetic resonance imaging is widely used as a main imaging technique for displaying the information of brain tumor.The magnetic resonance imaging(MRI)is a nondestructive three-dimensional information display technology,which can effectively display different types and sizes of brain tumors in different modes.This thesis mainly focuses on the research of brain tumor segmentation.Firstly,it introduces the background and significance of brain tumor segmentation,the development of brain tumor image segmentation technology and the current research status.The difficulties of brain tumor are summarized and the structure of this thesis is introduced.Detailing the problem of brain tumor segmentation and the latest research to solve technical means.1.MRI-based segmentation of brain tumor images has always been a hot research direction.My thesis from multi-mode MRI began to study.Firstly,fuzzy C-means algorithm and fast fuzzy C-means clustering algorithm based on histogram constraint are introduced.The level set algorithm and mixed level set algorithm are introduced.Fast fuzzy C-mean and mixed level set are two improved algorithms.In order to segment brain tumors effectively,an improved two-dimensional hybrid image segmentation algorithm for multi-mode brain tumors is proposed.The algorithm uses three modes of magnetic resonance imaging:TIC,T2 and FLAIR.The algorithm mainly uses the fast fuzzy C-means algorithm to cluster the fused image to segment the initial under-segmentation region;then the under-segmentation region uses the mixed-level algorithm to the final result of segmentation.2.Improved two-dimensional hybrid image segmentation algorithm for multi-mode brain tumors can effectively segment in large two-dimensional brain tumor in medical images,but the magnetic resonance imaging is three-dimensional image.At the same time,three-dimensional results are more conducive to clinical application.Therefore,this thesis improves the two-dimensional algorithm to multi-model brain tumor three-dimensional image hybrid segmentation algorithm,which is consistent with the two-dimensional algorithm in the flow.But the effect is better than the two-dimensional segmentation algorithm,and more conducive to clinical practice application.3.Due to the variety of brain tumor images and high complexity,traditional machine learning is better for image segmentation of the same type of features,but there is a clear deficit in brain tumor segmentation with large differences in features.Therefore,this thesis proposes a 3D brain tumor segmentation algorithm based on Voxresnet.The algorithm is based on the residual network to improve and achieve.A cascade net is used for multi-modal brain tumor images.In order to evaluate the effectiveness of the algorithm,three indicators:similarity coefficient,sensitivity and positive predictive rate is used to evaluate the results of brain tumor segmentation.The result of experiments show that the proposed algorithm has a good performance in brain tumor segmentation.This thesis study three different algorithms in different angles for brain tumor image segmentation algorithm,which have a certain effect.The article is summarized at the end.It is introduced that the limitations of this study and further needs to be done work,and the prospects for the application of brain tumors.