Coronary Artery Assisted Diagnosis And Treatment System Based On Deep Neural Network

In recent years,coronary artery-related diseases have become one of the main causes of disease death.With the development of China’s basic medical construction,the number of primary hospitals that need coronary intervention surgery has grown rapidly.However,there is a certain gap between the experience of primary surgeons and the doctors in the Tertiary Hospitals,and it is impossible to achieve a short-term and efficient individual diagnosis.With the rapid development of artificial intelligence,we have seen hope in AI-assisted diagnosis and treatment of coronary artery disease.This thesis is devoted to researching a real-time coronary artery assisted diagnosis and treatment system,which mainly completes the blood vessel extraction in angiography,eliminates the background noise generated during angiography,and provides a clear basic image for the diagnosis of various diseases.The main results of the paper are as follows.First,construct a coronary angiography data set for training and testing,where the data is labeled by relevant professional doctors.In these figures,the trend and outline of each blood vessel will be clearly marked.There are currently about 10,000 pieces of labeled data.In addition,fine labeling is carried out on the basis of coarsely labeled images.The annotated images can be directly used as training supervision for model training.Currently there are about 9,500 finely labeled images.Secondly,this paper experimented with a variety of convolutional neural network structures,and designed a neural network structure suitable for coronary angiography based on PspNet.Under the premise of ensuring speed,the model can fuse the coronary contour features Deep semantic features to accurately extract vascular information in coronary angiography images.Finally,on the basis of the blood vessel segmentation algorithm constructed in the previous section,deep optimization was performed to improve the segmentation efficiency and segmentation accuracy of the entire system.In the network training process,OHEM is introduced to solve the problems of sample imbalance and low training efficiency in the segmented network.In the network structure,the attention mechanism is applied to weight the context information of different regions,strengthen the detailed features such as blood vessel boundaries,and improve the final segmentation effect of the network.