Research Of Medical Image Segmentation Algorithm Based On Marginal Space Learning

Currently,the research on heart segmentation is mainly focus on the division of the atrium and ventricles.However,the cardiac surgery often relies on the whole structure of the heart.And a 3D model of the heart was reconstructed to assist doctors within the treatment of the heart.But it is difficult to obtain good segmentation results since the complexity of the heart structure and the noise produced by the beating of the heart.Therefore,the research heart segmentation has been attracted more and more attention.According to the high correlation between adjacent slices and different sequences in cardiac MRI image sequences,the heart segmentation algorithm is studied in the thesis and then proposed an automatic heart segmentation algorithm based on marginal space learning.The main research contents of this thesis include:Firstly,the traditional detection method based on Haar-like features and AdaBoost algorithm is researched.The Haar-Like features of the training sample images are extracted.And the target of heart can be automatically located by using AdaBoost algorithm and cascade classifier.Then,for solving the defect of the traditional detection method based on Haar-like features and AdaBoost algorithm,the detection algorithm based on boundary space learning is proposed.Experiments show our method can achieve the excellent performance.Finally,the heart segmentation based on Active Shape Model is used,and the average shape model is obtained by training the labeled heart region.Then,the heart model is initialized and the adaptive matching which based on the initial model and the heart boundary information of the detected image are performed.At last,the segmentation of heart image is achieved.The innovation in this thesis is that we propose the heart region of image based on the boundary space learning algorithm.Meanwhile,we compare the performance with the traditional detection method based on Haar-like features and AdaBoost algorithm.Then,the heart in the region is divided by the segmentation of full heart based on Active Shape Model.The experimental results show that the proposed scheme in this thesis can perform automatic and precise segmentation of the heart,which lays a foundation for the follow-up cardiac treatment and auxiliary diagnosis.