Research On Vessel Segmentation Of Coronary Angiography Image Based On Semi Supervised Learning

Vessel segmentation of coronary angiography image is the basis of clinical diagnosis of cardiovascular disease,so it is of great significance for vessel segmentation of coronary angiography image.Due to the lack of the label image data in coronary angiography,and the image segmentation based on semi supervised learning has become a research hotspot,this paper mainly studies the vessel segmentation methods of coronary angiography image based on semi supervised learning.Firstly,in view of the complex structure of coronary angiography image and the characteristics of a large amount of noise in the images,the CLAHE and NLM are used to preprocess the images,which improves the contrast between the background and blood vessels,suppresses the noise in the images and achieves the effect of image enhancement.Secondly,aiming at the problem of vessel segmentation in coronary angiography image,a method based on semi supervised adversarial model is proposed in this paper.The network framework takes a U-shaped network containing residual module and CBAM attention mechanism as the segmentation network.By adding the residual module,the problems of gradient disappearance and gradient explosion in the training process can be alleviated.Adding the attention mechanism module can solve the problem that the convolutional neural network is limited by the local receptive field,and the U-shaped structure can fuse the bottom features and the top features,so as to retain more detailed information.At the same time,by combining the adversarial learning,the segmentation network can learn the structural information from the feature distribution of the label image,so as to improve the segmentation performance.Through the analysis of experimental data,the effectiveness of this method is proved.Finally,aiming at the problem of vessel segmentation of coronary angiography image after cutting,a method based on semi supervised mean teacher model is proposed in this paper.This method includes both student model and teacher model,and the network structure of the two models is consistent.The teacher model does not participate in the model training,and its weight update adopts the stochastic weight averaging of the student model.In this network framework,the U-shaped structure with dense block is used as the network structure of the student model and teacher model.By adding the dense block to the network,the amount of calculation is reduced,the transmission of features is strengthened,and the segmentation accuracy and performance are improved.Through the analysis of experimental data,the effectiveness of this method is proved.