Research On Cardiac Image Segmentation Algorithm Based On Anatomical Knowledge

Cardiovascular disease is characterized by acute onset,low cure rate and high mortality rate.Therefore,clinicians attach great importance to the study of cardiovascular disease.With the help of clinical imaging technology,doctors can observe the internal anatomy of the patient and obtain more information for the diagnosis,thus improving the treatment effect.In recent years,image segmentation technology has attracted great attention because it can directly obtain the boundary of target tissue from complex medical images.And the segmentation results can guide doctors to develop treatment plans to improve the efficiency of diagnosis.This thesis combines the anatomy of the heart and masters its image features in medical images,mainly focuses on the segmentation of cardiac medical images.The specific work content and innovation are as follows:1.This thesis studies the commonly used imaging techniques and medical image preservation formats,and learns the anatomical knowledge of left ventricle and left atrial appendage,grasping the image features in medical images.2.The image contrast is improved by piecewise linear transformation and trough enhancement.And the anisotropic diffusion filter is used to remove the influence of image noise.These provide favorable image information for the next segmentation task.3.In terms of left ventricular segmentation.Firstly,in order to solve the initial sensitivity problem of the level set method,the heart image is pre-segmented by clustering algorithm,the left ventricular cavity is automatically extracted according to the anatomical features.And the initial level set function is constructed with the obtained binary image.Then,due to the structure of the papillary muscles on the endocardium,the evolution equation of the level set is improved to make the evolutionary process convex-preserving and improve the accuracy of the segmented results.Finally,according to the relative positional relationship and shape of the endocardium and epicardium,the convexity-preserving model is extended to a bi-level set model.The left ventricle endocardium and epicardium are segmented synchronously by using the constraints between 0 and K level set.4.In terms of left atrial appendage segmentation.Firstly,Otsu multi-threshold method is studied to remove background interference,so that a simple level set function can be used to segment the rich details of pericardium.Secondly,for the intensity inhomogeneity of images,the medical imaging model is applied to the image segmentation,and the energy function is designed by combining different scale local information to realize the accurate segmentation of the left atrial appendage.Finally,considering the prior knowledge that the gray value of the foreground is higher than the background in the image,this thesis forces the grayscale average of the foreground to be greater than that of the background during the calculation,thereby improving the stability of the algorithm.In this thesis,the performance of segmentation method is verified by public dataset and clinical dataset.A large number of comparative experiments show that the segmentation method is accurate and effective,and has certain clinical application value.