Research On MRI Image Segmentation Methods For Gliomas

As an important way of medical treatment,medical image processing technology plays an irreplaceable role.Brain glioma is one of the important reasons affecting human health.Segmentation and identification of glioma based on Magnetic Resonance Imaging(MRI)is one of the most effective medical diagnostic techniques at present.Therefore,the study on MRI image segmentation of glioma has important theoretical research and practical application value.The glioma MRI images have some special features,including complex and irregular shape,blurred outline,similargray values between targets and background.Brain tumor especially glioma MRI image segmentation problems are investigated in this dissertation based on theory of Nash equilibrium,Petri net based on Rough set and rough entropy,BoseEinstein condensation and Wormholeto explore new methods for improving the quality of medical image segmentation.The main research works are presented as follows.Study an improved support vector machine(SVM)based on Nash equilibrium theory for glioma image segmentation.Aiming at the problem that the setting of penalty parameters in SVM affects the accuracy of image segmentation,a new Nash equilibrium model with double constraints of entropy and standard deviation is proposed.The relationship between image characteristics and its segmentation process,Nash equilibrium theory and its reasoning mechanism is studied.The parameter calculation method of the new Nash equilibrium model is obtained,and the Nash equilibrium process is constructed.An improved SVM based on the new Nash equilibrium model is proposed.The penalty parameters in SVM are determined by Nash equilibrium reasoning with double constraints of entropy and standard deviation.Experiments on brain glioma MRI image segmentation are carried out to verify the effectiveness of the proposed method.The clustering method of target edge region based on Nash equilibrium theory is studied.A two-step Nash equilibrium clustering method is proposed to solve the problem of blurring the edge regions of glioma.The target and background regions of glioma are obtained by the maximum similarity judgement(the maximum similarity between the nodes in the target region)and the minimum similarity judgement(the minimum similarity between the nodes in the target region and the background region).Based on Nash Equilibrium Theory,the C-V model is improved to obtain the contour of glioma.To solve the problem of similarity of the edge regions of brain tumors,a texture similarity region judgment and merging method based on Nash Equilibrium is proposed.After obtaining the target regions and background regions of brain tumors,the C-V model of multi-texture features is improved based on Nash Equilibrium Theory,and the node features in the image are mapped to the C-V model.The average gray parameters of C-V model are set by Nash equilibrium reasoning.The target contour of brain tumors was obtained by the improved C-V model.The effect of this method is verified by the experiment of glioma image segmentation on MRI.A method of Petri net image segmentation based on rough set and rough entropy is studied.Aiming at the problem of inaccurate contour line of brain tumor image segmentation caused by general image segmentation which only judges one node on the contour line but not the adjacent nodes on the contour line,a method of brain tumor image segmentation based on Petri net of rough set and rough entropy is proposed.Two stages of rough segmentation and fine segmentation are proposed: the first stage is rough segmentation based on rough set and rough entropy to obtain the initial contour of the target object;the second stage is fine segmentation based on Petri net,which uses Petri net to select the multilateral boundary and forward and backward correction to get more accurate target contour.Experiments verify the effectiveness of this method in improving the accuracy of image segmentation.The model of glioma image segmentation based on Bose-Einstein Condensate(BEC)is studied.Aiming at the problem that the shape of glioma is usually cystic or ring-enhanced edge contour,which makes it difficult to segment its image accurately,a new method of medical image segmentation is explored.In this paper,try to apply the dynamic process of phase transformation and stability theory of BEC theory to glioma image segmentation.The kernel function of support vector machine(SVM)is constructed based on BEC theory.A model and image segmentation method of SVM glioma image segmentation based on BEC kernel function are proposed.Through different types of brain glioma image segmentation experiments,the segmentation effect of this method is compared with other similar methods.A method of glioma image segmentation based on quantum and wormhole-behaved particle swarm optimization(QWPSO)algorithm is studied.To solve the problem of brain tumor image segmentation with complex shapes such as "bottleneck" and "hard tail",the quantum and wormhole theory is introduced to improve the quantum particle swarm optimization(QPSO),which is called Quantum and Wormhole-behaved Particle Swarm Optimization(QWPSO).The nodes are divided into seed particle nodes and pixel particle nodes.The classification formulas for distinguishing the two kinds of nodes are given.The wormhole hyperbolic path formula and the QWPSO algorithm formula are proposed.QWPSO method explores a new way to solve the problem of segmentation of complex shape brain tumors in MRI images.The comparative effect of this method and other methods are verified by experiments.