To Explore The Risk Factors Of Recurrent Cardiovascular Adverse Events In Postoperative Patients With Coronary Heart Disease Under The Guidance Of Deep Learning

Research background and purpose:Coronary heart disease is one of the major diseases threatening human health.Some patients with coronary heart disease will have cardiovascular adverse events again after interventional therapy.Intravascular plaque rupture,platelet aggregation,thrombosis leads to the occurrence of ACS.The stability of atherosclerotic plaques is directly related to the prognosis of patients with ACS.Therefore,early recognition and quantification of corresponding plaques in OCT images are particularly important.At present,the popularity rate of cardiovascular OCT in Chinese hospitals is not high.Many physicians have insufficient understanding of coronary OCT images,and there is a large subjectivity in interpretation.At present,the recognition of coronary artery OCT mainly relies on manual recognition and labeling,and hundreds of images are generated in one OCT regression imaging,which greatly increases the time cost.There is an urgent need to apply artificial intelligence automated recognition to coronary OCT to save time and cost,improve marker accuracy and reduce subjective dependence.Deep learning is an efficient and highly accurate machine learning algorithm.At present,some studies have applied it in medical imaging,but there are few studies on its application in cardiovascular OCT.At present,there is still a lack of studies on the automatic analysis of pathological information in OCT images of patients with coronary heart disease by deep learning.The purpose of this study was to preliminarily explore the accuracy of deep learning in automatic identification of lesion information in coronary OCT images of patients with coronary heart disease,and to explore risk factors of recurrent cardiovascular adverse events after PCI of coronary heart disease on the basis of combining deep learning,so as to guide clinical intervention in advance.To preliminarily explore the feasibility of the clinical application of deep learning technology in the cardiovascular field,and lay a solid foundation for the extensive clinical application of deep learning in coronary OCT in the future.Materials and Methods:In the early stage,a large number of cardiovascular OCT images were manually labeled,and U-net deep learning was applied to the programmed learning of the manually correctly labeled OCT images,so that they could be applied to the automatic recognition of vascular lesions in cardiovascular OCT images.Subsequently,400 OCT images were divided into 8 stages,with 50 OCT images in each stage.The junior physicians who first encountered OCT images were analyzed,and the images were analyzed using deep learning.Finally,senior physicians made judgments,and the time and accuracy of the two at each stage were recorded.Clinical data and OCT images of 150 patients with coronary heart disease admitted to the cardiovascular Department of Sichuan Provincial People’s Hospital from June 2021 to December 2021 were retrospectively collected and analyzed.They were divided into event group and non-event group according to whether major adverse cardiovascular events occurred during the follow-up period.The number of types of intravascular lesions depicted by OCT was rapidly analyzed by deep learning.Basic clinical information,laboratory indicators and OCT images were compared between the two groups.Multivariate Logistic regression analysis was used to determine the independent risk factors for MACEs in patients with CHD after PCI.Result:1.In the previous stage,it was found that the accuracy and time of the eight stages of deep learning were relatively stable,while the accuracy of each stage of manual film reading by junior physicians showed an upward trend and the time gradually decreased.From stage 1 to stage 3,the accuracy of deep learning reading was higher than that of manual reading(P<0.05).The accuracy of manual reading tended to be stable in the four stages,and there was no statistical significance in the accuracy between deep learning and manual reading in the subsequent stages(P>0.05),but the accuracy of deep learning is always better than that of manual reading.Moreover,in terms of the time spent on reading films at each stage,the time spent on deep learning software is significantly less than that of manual reading films..2.There was no significant difference in basic information between the two groups(P> 0.05),however,in terms of diabetes prevalence and estimated glomerular filtration rate abnormalities,the number of patients in the event group was significantly higher than that in the non-event group(P<0.05).By comparing the biochemical indexes of the two groups,there was no significant difference in HDL cholesterol and triglyceride between the two groups(P> 0.05).However,the levels of LDL cholesterol and total cholesterol in the event group were significantly higher than those in the non-event group,and the difference between the two groups was statistically significant(P<0.05).Cardiovascular OCT images of each patient were analyzed by deep learning and the number of each type of intravascular lesion was obtained.There were no statistically significant differences in the number of fibrous plaques,calcified plaques,and cavities between the two groups(P>0.05).However,the number of macrophages and thin fibrous cap plaques in the event group was significantly higher than that in the non-event group(P< 0.05).Multiple Logistic regression analysis showed that age,diabetes prevalence,renal insufficiency,thin fiber cap plaque and macrophage were risk factors for MACEs.Conclusion:This study found that deep learning has the characteristics of rapid and relatively accurate lesion recognition in cardiovascular OCT images,and its preliminary application in coronary OCT images has certain feasibility.Multivariate Logistic regression analysis was used to determine that age,diabetes,renal insufficiency,thin fibrocap plaque and macrophages were independent influencing factors for the occurrence of MACEs,and age had the largest area under ROC curve in predicting the risk of MACEs(AUC=0.708),followed by renal insufficiency(AUC=0.664).Finally,thin fibrous cap plaque(AUC=0.648).This helps to identify high-risk groups early and provides a theoretical basis for early intervention.