The Research On The Segmentation Method Of Echocardiography

Because of the portability and low cost of ultrasonic imaging equipment,echocardiography has become the first choice for the visualization of the heart.For the improvement of the quality of ultrasound images and the accurate segmentation of the heart,the core problem is to improve the effectiveness and accuracy of clinical diagnosis.The method of manual segmentation is used in clinic,which is time-consuming and tedious,and the segmentation results are affected by subjective factors.Therefore,the automatic or semi-automatic segmentation of echocardiography has become the focus of research in this field.Based on the support of the National Natural Science Foundation of China(61471124),the accuracy and reliability of ultrasonic image segmentation with high noise and low contrast are discussed in this paper.We propose an improved bilateral filtering denoising algorithm,a new algorithm of left ventricle segmentation based on variable shape restriction model,and a contour detection algorithm based on depth convolution neural network.Specific research contents include the following aspects:Firstly,this paper proposes an improved bilateral filtering algorithm to reduce noise of echocardiography.The algorithm for edge information loss problem,introduce the edge detector and the similarity function,combining spatial proximity and similarity of pixels on the edge of a compromise,realize the protection and better able to remove noise.The experimental results show that the proposed algorithm has better noise reduction effect and stronger ability to protect the boundary information.In addition,by comparing the results before and after filtering,it is found that the accuracy of image segmentation is higher after denoising.Secondly,this paper designs a variable shape restriction model and proposes a new algorithm for automatic segmentation of the left ventricle.The algorithm uses fuzzy C means clustering to segment the global image,and the initial segmentation result is used as the control parameter of the level set iteration.Then the level set method is used to segment the ultrasonic image second times,which can further reduce the influence of noise and get the relatively clear contour.In the end,we designed an ellipse like contour force field as a priori model for the left ventricular structure.Experimental results show that the proposed algorithm is more accurate and robust.And the average error rate of the region overlapping in image sequence segmentation is as low as 2.96%,and the average difference value is only 2.38%.Thirdly,in this paper,a convolutional neural network model is designed and the depth features are studied in order to realize the contour detection of echocardiography.In this part,the neural network is used as a black box for feature extraction,and a special training strategy is designed.We divide the boundary information into several subclasses,and then use different model parameters to fit each sub class.At the same time,the loss function is improved,and the loss of a positive sample is assigned to each sample subclass,and then lears the parameters.The experimental results show that the proposed method is able to learn more recognition features and the segmentation results are very similar to those of manual markers.Compared with the SE algorithm,the average accuracy of the proposed algorithm is improved by 4%.In order to solve the problem of boundary leakage and low accuracy in ultrasonic image segmentation,this paper proposes three algorithms and improved scheme.The experimental results show that the performance of the proposed segmentation algorithm has some advantages and has high practical value.This research lays a better foundation for the following auxiliary diagnosis.