Research On Key Algorithms In Coronary Stenosis Recognition And Analysis Based On CTA Image

In recent years,a number of medical research reports have pointed out that Cardiovascular Diseases(CVD)account for more than 40% of the deaths of urban and rural residents in China,and it has become one of the main Diseases endangering the life safety of residents.Coronary artery disease(CAD)is more harmful than other cardiovascular diseases.Therefore,early diagnosis of coronary artery disease has important clinical value.CCTA is widely used in the diagnosis of coronary artery disease.By analyzing CCTA images,the health status of coronary arteries can be analyzed,such as whether there is stenosis or not.The automatic screening of CCTA depends on the high-precision segmentation of coronary artery lumen,but it is difficult to rely on manual segmentation of coronary arteries.The detection and quantification of stenosis mainly rely on doctors to read the original CCTA image,which is not only not intuitive,but also time-consuming and labor-intensive.Therefore,automatic vascular segmentation method is needed.The automatic detection and quantification method of stenosis based on deep learning is mainly studied in this thesis.This thesis investigates the existing coronary artery segmentation algorithms and stenosis detection algorithms.Several mainstream algorithms which are widely used are summarized.Traditional algorithms require complex processing,parameter adjustment and other steps in application.In recent years,with the continuous development of deep learning,it has been widely used in medical images.Therefore,based on the excellent algorithms,this thesis puts forward a new idea of coronary artery segmentation and stenosis detection.The main work of this thesis is as follows:Aiming at the task of coronary artery segmentation,in order to obtain high-precision segmentation,this thesis extends from the traditional active contour model algorithm and combines with deep learning to propose a two-dimensional depth iterative deformable network(2D DIDNet).The segmentation and reconstruction of blood vessels are carried out through data sampling,initial contour setting,iterative deformation,3D reconstruction and so on.At the same time,in order to obtain more continuous and smooth segmentation results,a threedimensional graph convolution network(3D GCN)is designed from the perspective of graph convolution neural network.In this method,the blood vessel is regarded as a three-dimensional tubular structure,and the position of each contour point is predicted by graph convolution.Finally,the segmentation result with sub-pixel accuracy is obtained.In addition,3D and Pseudo3 D depth iterative deformation network(3D DIDNet)is designed for comparison.Aiming at the task of detection and quantification of coronary artery stenosis,in order to facilitate the comparison with the visual diagnosis of doctors,this thesis detects and locates the stenosis on the longitudinal section of blood vessels in CMPR space.From a new perspective,this thesis regards this research as a target detection task,draws lessons from the transformer model which is excellent in the field of natural language processing,and proposes a narrow detection and quantization model(SFPT)based on transformer combined with residual network and feature pyramid network.This method avoids the complex steps of preprocessing and postprocessing in the general detection algorithm.In order to meet the speed and accuracy requirements of clinical application,this thesis improves the commonly used target detection model,integrates the attention mechanism and mixed normalization layer,and proposes a stenosis detection and quantification model(AMDNet)based on attention mechanism and mixed normalization,which improves the detection performance of the algorithm without increasing the amount of calculation and prediction time.At the same time,system development is also carried out in this thesis,and the software interface for displaying segmentation and detection algorithms is designed through GUI programming.Through the system interface,it is convenient to call relevant algorithms and display algorithm results.The results of this thesis show that the proposed algorithm achieves good results on experimental data and has the value of space expansion and application.Finally,this thesis summarizes the work done in this thesis and analyzes the prospect of the future work based on deep learning knowledge.