Research On Risk Prediction Of Cardiovascular Surgery Indicators Using Retinal Image Classification Algorithm Based On Deep Learning

The vascular features in retinal images can reflect the severity of cardiovascular disease and indirectly reflect the risk of cardiovascular surgery to a certain extent.Because retinal image has the advantages of low cost and noninvasive to human body,it is an important way for clinicians to analyze and evaluate cardiovascular disease by using retinal image to predict the risk level of cardiovascular surgery indicators.In recent years,deep learning has gradually become the mainstream means of processing retinal image classification tasks.However,most of the current research work mainly through optimizing the structure of the model to learn the feature information of the input image,without considering the relationship between the images.On the other hand,these algorithms are mainly based on the diagnosis of ophthalmic diseases,which need to take into account the biological characteristics outside the blood vessels in the retinal image,such as leakage,microaneurysm,optic disc and so on.However,for cardiovascular disease,these features are interference items,so the algorithm is required to further filter these invalid features.In addition,most of the algorithms lack a certain interpretability mechanism,which is not conducive for clinicians to investigate the key areas of retinal images.In view of the shortcomings of the existing retinal image classification algorithms in predicting the risk of cardiovascular surgery indicators,this paper proposes two risk classification algorithms based on deep learning from different perspectives.The idea of the first method is to use the vascular segmentation model to extract the vascular image of the retinal image,and then use the double branch classification algorithm based on deep learning to consider the feature information of the input image and the relationship information between the images for classification and prediction.The second idea is to directly extract the middle layer parameters of the vascular segmentation model as the feature input of the subsequent classifier.Clinicians can judge the reliability of the input features by observing the vascular map generated by the model,which greatly enhances the interpretability of the model.In addition,attention heat map can be generated by two methods to help medical staff understand the key areas of the model.In this paper,two groups of real data sets from Guangdong Provincial People’s hospital are used respectively.The experimental results show that the two algorithms proposed in this paper are effective and robust in the risk classification task of retinal image surgery indicators.