Research On Coronary Vessel Extraction And Intracluminal Plaque Labeling Algorithm In CTA Image

According to the survey,the heart disease and stroke are the main causes of death in middle-aged people,and the mortality rate of them ranks first and second respectively in the global disease mortality rate.With the development of economy,the number of patients with cardiovascular diseases continues to increase,and the morbidity and mortality rate are in a rapid rise stage.With the development of medical imaging technology,doctors can preliminarily analyze and diagnose the situation of cardiovascular diseases through patient’s images.Within these medical imaging technologies,CT Angiography(Computed Tomography Angiography,CTA)imaging technology because of the popularity of imaging equipment,high imaging resolution,non-invasiveness,simple operation and low cost has gradually become the main method for coronary heart disease diagnosis.In this paper,the CTA images of patients were selected as the object to study the algorithm of coronary vessel extraction and luminal calcified plaque labeling.The main research was as follows:1.In the study of coronary artery extraction,this paper proposes a new improved MOSSE(Minimun Output Sum of Squared Error filter)tracking algorithm based on feature fusion.The proposed algorithm uses the fusion of vascular geometric feature and similar feature to ensure the vascular location and then uses the MOSSE method for tracing.It solved the problems of traditional MOSSE method such as semi-automatic,tracking only a single target and cannot tracking dramatically changed target.It realized automatic and efficient tracing of coronary arteries.Through the analysis of experimental data,the accuracy of the proposed algorithm in coronary vessel extraction can reach 94.3%,which was better than other tracking methods and prove the advantages of the proposed algorithm.2.In the label and recognition task of lumen calcified plaque,this paper proposes a new network model named DSE-U-Net which based on attentional mechanism and multi-scale information fusion module(U-Net with Squeeze-and Excitation and Semantic-Embedding module,U-Net with Double SE module).The new network enhanced the dependence between convolutional feature channels and also added the multi-scale information which improved the over-segmentation and low-precision problems of traditional U-Net network and realized automatic labeling of lumen calcified plaques.We evaluate the DSE-U-NET model and the traditional U-NET model in two ways,one is Dice and another is Miou,and both of them proved that the model proposed in this paper has more validity.3.This paper analyses the existing 3D visualization methods,including volume rendering and surface rendering,and then uses the 3D visualization tool vtk combined with the MC(Marching Cubes)algorithm which in surface rendering to realize 3D visualization of coronary vessels and the automatic labeling of lumen calcified plaque,so as to help doctors diagnose and treat more intuitively.