Research On Corpus Callosum Segmentation And Fiber Bundle Tracking In Dti Image

Corpus callosum is the fiber bundle connecting the corresponding parts of the left and right brain,making the brain become a whole in function.For CT images reconstructed by computed tomography(CT)and MR images obtained by conventional magnetic resonance imaging(MRI),only the general shape of corpus callosum can be obtained by direct segmentation in the images,which can not be used for corpus callosum related pathological analysis or experimental research,and it is difficult to reconstruct the tiny fibers in corpus callosum.In the diffusion tensor imaging(DTI)image,the shape of the corpus callosum is segmented according to the different degrees of anisotropic diffusion of the fiber bundles in the corpus callosum and the surrounding tissues.Then,the fibers in corpus callosum were tracked according to the diffusion characteristics of water molecules in voxels,and the three-dimensional reconstruction of the corpus callosum is realized.In this way,it can not only provide specific diagnosis basis for corpus callosum lesions,but also provide surgical navigation during the operation,and also timely check the recovery of the patient after the operation.In this thesis,we focus on the problem of difficult to determine the direction of voxel fibers in the process of fiber tracking and the problem of over segmentation in the process of watershed segmentation of corpus callosum,the main research of this thesis is as follows:1.Research on brain fiber tracking method based on K-means and moving least squares(MLS).The existing deterministic fiber tracking algorithm only considers the tensor of the voxel itself,which leads to large errors in the final tracking results.Therefore,we propose a deterministic fiber tracing method KM-STT that considers both the voxel tensor ellipsoid characteristics and the neighborhood voxels information.Firstly,the voxel tensor is modeled by an ellipsoid,and then the voxels are clustered by K-means according to the difference of the ellipsoid model;secondly,the adaptive anisotropic Gaussian function is constructed and used as the weight function of MLS,for the voxels whose tensor models are disk-shaped ellipsoid and spherical,MLS fitting is used to update their tensor values;finally,in the new tensor information field,streamline tracing is used,tracking algorithm performs fiber tracking to obtain a more complete and accurate brain nerve fiber bundle.A simulated human brain data set and a real clinical data set are used to qualitatively and quantitatively evaluate the tracking results.The experimental results show that in the real clinical data,KM-STT can track more complete fiber bundles than STT;In simulated human brain data,the KM-STT algorithm can track up to 23 correct fiber bundles,the correct fiber connection ratio reaches 45%,the wrong fiber bundles are reduced by 16 bundles compared with UKF,and the wrong fiber connection ratio is reduced to 28%.2.Research on the method of corpus callosum segmentation based on improved marker watershed.In the FA image,the anisotropic diffusion characteristics of voxels in the same tissue are similar,and the anisotropic diffusion characteristics of voxels in different tissues are quite different.Therefore,the corpus callosum can be segmented according to the anisotropic diffusion characteristics of the voxels,and the segmented region is used as the region of interest(ROI)for the subsequent 3D reconstruction of the corpus callosum.Watershed algorithm has high precision and high speed,but it has over-segmentation problem when directly using watershed algorithm to segment corpus callosum in FA image.Therefore,we apply genetic algorithm,extended minimum transformation and forced minimum marker to watershed algorithm,and propose an improved watershed segmentation method based on marker.Firstly,FA image is calculated according to DTI image data,and the gradient image of FA image is calculated by Sobel operator.Secondly,the adaptive genetic algorithm is defined,and the inter class variance function is taken as the fitness function of the genetic algorithm to iteratively obtain the optimal solution;then,the optimal solution is taken as the threshold of the extended minimum transformation,and the region minimum is transformed to obtain the marked image,and according to the labeled image,the local minimum value area of the gradient image is preserved;finally,watershed algorithm is used to segment the corpus callosum from the gradient image after updating the local minimum region.This method can effectively solve the problem of excessive segmentation in the use of watershed algorithm,obtain better results of corpus callosum segmentation and shorten the segmentation time.