The Generating Model Constraints Graph Cuts Label Fusion Algorithm And Research On Three-Dimensional Visualization

Human brain contains many important structures,such as hippocampus,cerebellum,amygdala,brainstem and caudate nucleus.As one of the most important structures in the human brain,the hippocampus’s structural abnormalities and lesions can lead to Alzheimer’s Disease,major depression and other mental diseases.Therefore,using image segmentation algorithm to achieve accurate segmentation of human brain hippocampus can provide important guidance for clinical diagnosis.In this paper,the human brain Magnetic Resonance image is used to study the multi-atlas label fusion algorithm,so as to realize the segmentation of human brain hippocampus structure,and the segmentation results were applied to the three-dimensional visualization part,to realize the three-dimensional reconstruction of the human brain hippocampus mechanism,and to display the segmentation results intuitively,providing a basis for clinical diagnosis and medical research.Human brain Magnetic Resonance images are extremely similar in gray scale between different brain tissues,and the tissue boundaries between different structures are not obvious.In addition,there are errors in the manually labeled images by experts,which makes the automatic segmentation of human brain Magnetic Resonance images very difficult.When the difference between the image to be segmented and the reference image is large,the traditional segmentation algorithm using a single atlas is likely to cause serious segmentation errors.In order to improve the accuracy of segmentation results,this paper adopts atlas set instead of single atlas registration,that is,a series of atlases are used to register the target image respectively,and finally the segmentation results of multiple atlas are fused to obtain consistent segmentation results,so as to achieve accurate segmentation of human hippocampal structure.In the phase of atlas pre-selection,gradient information and mutual information are used to screen out highly similar atlas images,which can reduce the impact of false labels on the accuracy of label fusion and improve the segmentation accuracy of human brain magnetic resonance images.Before multi-atlas registration,the hippocampal-centered Magnetic Resonance image was extracted using the bounding box algorithm,and the extracted image block was used to replace the original image for registration and fusion,so as to improve the time-consuming and high computational complexity of the traditional method for the whole image processing.In the label fusion stage,by studying the traditional segmentation methods of human brain Magnetic Resonance,we master the segmentation principle of each algorithm and the shortcomings in the segmentation process.We proposed a Graph Cuts label fusion method using Generative Model constraints.The Generative Model can effectively representing the spatial correlation and local statistical characteristics of image pixels,the Graph Cuts algorithm was used to optimize the energy function rapidly and accurately segment the brain hippocampus structure.Compared with the traditional label fusion method,the proposed algorithm has higher segmentation accuracy.Finally,in the stage of medical image three-dimensional visualization,this paper deeply studied two commonly used three-dimensional visualization algorithms:surface rendering and volume rendering,studied the principle of surface rendering and volume rendering algorithms,and realized the three-dimensional reconstruction of medical image with the VTK visualization toolkit and QT framework.And,by using the scroll bar to adjust the transfer parameters in real time,the visualization of three-dimensional image is realized.