Prediction Of Occlusive Branch Of CTO Pathological Contrast Based On Deep Learning

In recent years,diseases caused by coronary artery have become an urgent threat in the cardiovascular field due to their complex causes and various complications.One of the most difficult conditions for diagnosis and surgical intervention is Chronic Total Occlusion(CTO),which is an occlusion of coronary branches.At present,high-quality coronary angiography data are still scarce,and coronary arteries are characterized by complex branch structure,etc.,so there are few studies in the field of CTO branch prediction in academia and industry.Therefore,this thesis presents a scheme to predict the branch direction of CTO angiography occlusion by combining computer vision technology.The main research contents and achievements of this thesis are as follows:First,the coronary artery data set is independently established according to the explored feasible scheme,including data desensitization,data(crude/fine)annotation,data grouping according to the extended form,etc.,and a pixel-level data enhancement algorithm is designed.Second,on the basis of this data set,the deep learning method is used to predict the direction of blocked branches,explore the traditional and deep learningbased methods of medical blood vessel prediction generation,and focus on the analysis of the advantages of anti-generation network.An innovative strategy of catheter alignment and Electrocardiography(ECG)denoising is proposed to achieve better data alignment.In addition,this thesis introduces a spatial-channel mixed attention mechanism to better achieve local feature extraction in vascular branch prediction.Thirdly,a quantitative evaluation method for angiogenesis effect was proposed.Firstly,skeleton/center lines of the predicted results were extracted,and then morphological similarity scores were calculated.The feasibility and effectiveness of the proposed scheme and improvement are verified by relevant experiments.