Study Of Cardiac Image Segmentation Algorithm Based On Graph Cuts

With the development of economic life,the threat of cardiovascular disease to human health is more and more serious.In clinical practice,fast and efficient diagnosis through medical imaging techniques for the prevention and treatment of heart disease has important significance.Currently,heart disease diagnosis depends on manual segmentation of cardiac images for various structures by experienced doctors in order to get accurate anatomical information and assess cardiac function.This is a very time-consuming task in practice.In this paper,we propose the Graph cuts algorithm integrating multi-dimensional features to study the cardiac image segmentation.Firstly,the basic theory of graph is described in detail.And then,two classical segmentation algorithm based on Graph Theory are given: Grab cut and Graph cuts.Afterwards we carry out experimental analysis of these two methods and display their advantages and disadvantages.Finally,we choose Graph cuts as the basis algorithm of this paper.Since the right ventricular wall is thin and tissue contrast is low,the accuracy of traditional segmentation is not high.In order to solve this problem,we use the edge feature to replace the original cardiac MRI images.In this paper,the right ventricular segmentation was achieved by combining the Gabor and HOG features.Experimental results show that the proposed method is effective to improve the accuracy of segmentation results.On the basis of previous algorithm,we achieve atlas-guided segmentation with prior knowledge for the aorta and left atrium.At the first,the corresponding structures are segmented by previous algorithm.They are considered as the prior information,which added into the registration framework.Here the kappa statistic combing mutual information for similarity measure,the affine transform for global transformation,local transformation using B-spline FFD model,and adaptive gradient descent algorithm are chosen to obtain the best results.Experimental results show that this segmentation method is more robust and effective.