Research On Aortic Segmentation And 3D Reconstruction Based On Improved UNet

The aorta is the main conduit for blood to all parts of the body and the thickest artery in the body.The incidence of aortic disease is increasing year by year in China,and aortic disease is fierce,which seriously threatens people's life safety.Among them,aortic dissection is the most common and dangerous kind of acute and severe disease in clinic.Patients often come to hospital because of chest pain and abdominal pain.If the diagnosis and treatment is not timely or the patients and their families do not realize the danger of procrastination,the mortality rate is very high.There are many similar organ tissues around the aorta,and for attending doctors,they need more effective information,so accurate segmentation and extraction of aortic regions is particularly critical.Considering the practical factors,the thesis designs a aortic segmentation method based on improved UNet,which can accurately segment the aortic region and realize three-dimensional display,which brings great convenience to the doctor's diagnosis.The main work of this thesis is as follows:(1)Through cooperation with Tianjin chest Hospital,the image data set of human thoracic and abdominal aorta CT was constructed.The aortic area was accurately marked by doctors,and then the CT image was preprocessed.The noise on the image was removed by filtering algorithm.To avoid the loss of some important information,the image is sharpened by Laplacian operator to highlight the edge features of each tissue.Finally,the CT image aortic region is enhanced by Frangi2D filtering algorithm based on Hessian matrix,which lays a foundation for subsequent aortic segmentation and extraction.(2)The traditional UNet model is improved according to the shape characteristics of the aorta and the Attention-UNet model,Atrous-UNet model and improved Res-UNet model suitable for two-dimensional aortic segmentation are designed.Comparing the results of the three models training with the gold standard,it is concluded that the improved Res-UNet network model designed in this thesis,which has high accuracy for aortic segmentation and can realize accurate segmentation and extraction of aortic region in human chest and abdominal cavity CT image.(3)Through the three-dimensional display of the aortic wall,it is convenient for doctors to observe the thickness information of the aortic wall.Based on the VTK and MFC development environment,a three-dimensional visualization system of the aorta is designed by using the mobile cube algorithm.The three-dimensional reconstruction of the aortic segmentation results extracted by the method designed in this thesis can accurately display the three-dimensional shape of the aortic model and visualize the related operations.Aortic segmentation method based on the improved UNet network model is designed to fully consider the shape characteristics of the aorta and can efficiently and accurately segment the aorta.The average overlap rate of the experimental results is 15.85% higher than that of the existing methods.Therefore,the accuracy of the aortic segmentation algorithm proposed in this thesis is basically in line with the practical application requirements,and the reconstructed three-dimensional aortic model shows better results.