The Detection Of Coronary Artery Disease Based On Deep Learning Technology

Coronary atherosclerotic heart disease has become one of the main diseases that endangers the health of Chinese residents,and research on prevention and diagnosis of this disease has become an urgent need.With the rapid development of interventional diagnosis and treatment of coronary heart disease in China,the relevant imaging technology has gradually matured.Clinically,detecting the area of interest or lesions in medical images is a key step in the diagnosis of coronary heart disease.However,in the actual interpretation of medical images,the accuracy and efficiency of diagnosis need to be improved.Computer-aided diagnosis is a technology that assists doctors in analyzing and interpreting medical images.With the development of artificial intelligence,it has become a trend to integrate artificial intelligence technology into computer-aided diagnosis technology.Based on the background described above,this thesis carries out research on coronary artery disease detection algorithms based on deep learning,and gradually optimizes the system.The main work is as follows:The dataset establishment:The thesis first identified the coronary angiography image as the clinical diagnosis basis.Due to the lack of high-quality annotated coronary angiography image disease datasets,a dataset was created.According to the requirements of the detection task,the process standard for the data set establishment was developed.Secondly,the professional doctors were coordinated to label the data sets,Then the data is reviewed for reliability,desensitized,etc..Finally,the data set of coronary artery lesions consistent with this study was established.Multi-scale model design and optimization:After summarizing the research on common deep learning-based target detection networks,this thesis proposes a coronary artery disease detection method based on Resnet101-Faster RCNN and adds a pyramid feature map module on this basis.The multi-scale feature with pyramid features improves the ability to detect multi-scale lesions in coronary angiography images In order to eliminate the influence of the extremely uneven foreground background that is commonly found in target detection,we use the Focal Loss loss function to optimize the foreground background classifier in the regional proposal network,so that the network pays more attention to difficult samples,so as to obtain higher quality proposal of the detection frame improves the average accuracy of the detection network.At last,ROIAlign is applied as a processing layer that maps candidate frames to feature maps and outputs uniform size features,which further improves the network’s ability to detect small-scale lesions.In this thesis,a performance test experiment is performed for each optimization in the system design,and the experimental results are analyzed.The experimental results prove the effectiveness of the method described in detecting coronary artery disease.