| Background:Coronary heart disease is the leading cause of death in the world,which seriously threatens human health.The main pathological basis of the disease is the occurrence and progression of atherosclerosis of coronary artery.Plaques in the lesion may cause the stenosis and obstruction of coronary artery.When the vascular stenosis progresses to a certain extent,it may induce myocardial ischemia and hypoxia,which may lead to acute myocardial infarction or even sudden death.Early prediction and timely intervention of coronary heart disease may help to prevent the occurrence and delay the progression of coronary heart disease,reducing the death and the treatment cost caused by coronary heart disease.In the screening and diagnosis of coronary heart disease,coronary CTA has always played the role of gatekeeper of coronary angiography because of its noninvasive,repeatability,relatively cheap price and high negative predictive value.The results of the DISCHARGE published in the New England Journal in April 2022 also confirmed the advantages of coronary CTA as a screening test for patients with chest pain: CTA and coronary angiography were used as the initial diagnostic imaging strategy to guide the treatment of patients.The incidence of mace events was similar between the groups,while the incidence of surgery related complications was lower.According to the expert consensus of SCCT2021 coronary artery CT angiography,the gold standard for the diagnosis of ischemia has evolved from the percentage of vessel diameter stenosis obtained by invasive coronary angiography to a physiological measurement that can reflect coronary blood flow and invasive blood flow reserve fraction which induces ischemia.Coronary CT angiography or invasive coronary angiography can determine the degree of vascular structural stenosis,but not the functional ischemia.The FFR obtained from conventional CCTA image data by computational fluid dynamics can make it possible to comprehensively evaluate the anatomical structure and physiological function of coronary artery noninvasively.Moreover,CTA information can be transmitted remotely,so CT-FFR can carry out network diagnosis.Because of the above advantages,CT-FFR can replace invasive FFR detection in some fields.Compared with structural coronary angiography or invasive FFR functional examination,CT-FFR may have a better performance in guiding the formulation of treatment strategies,improving prognosis and reducing medical costs.The occurrence of cardiovascular accidents of coronary heart disease is related to many factors.The establishment of prediction models for cardiovascular events in coronary heart disease based on the nature and characteristics of plaques has always been a hotspot.However,the data of most research are based on the information of a certain time cutoff point,which has certain limitations.This experiment intends to analyze the coronary CT information at twotime points through multi-dimensional analysis,and apply the method of machine learning to obtain the structural characteristics of coronary vessels,the nature of plaques,the baseline value and follow-up value of hemodynamics,establish early warning models by various methods,and screen the dominant models.Objective:Based on the cohort of middle-aged and elderly patients with coronary heart disease or suspected coronary heart disease,CTA image follow-up data such as plaque structure stenosis,nature and hemodynamics were obtained.Through machine learning technology and multidimensional data fusion,the imaging prediction model of ACS events is established to improve the prediction accuracy.Methods:1.Retrospective analysis was performed on consecutive hospitalized patients in the second and fourth medical centers of the PLA General Hospital from January 1,2015 to December31,2019.Among the 452 patients with documented acute coronary syndrome,101 patients who underwent CCTA more than once within 3 years were selected as ACS group.The other101 patients without ACS events with the propensity score were selected as a case control group.2.The hemodynamics of coronary artery was calculated and analyzed by deep-learning based CT-FFR.3.The properties of coronary plaque were quantitatively analyzed by QAngio CT software.4.The characteristics of vascular anatomy,plaque nature,baseline value,follow-up value and change value of CT-FFR data were analyzed.5.Using the method of logistic regression,the vascular anatomical structure,plaque nature,baseline value,follow-up value and change value of CT-FFR were included respectively,the ACS early warning model was established,and the area AUC under the curve was calculated.6.Using the method of logistic regression,the data of all 27 parameters: vascular anatomical structure,plaque nature,baseline value,follow-up value and change value of CTFFR were included to establish ACS early warning model and calculate AUC.7.The important factors are screened by the method of logistics.Taking the five factors with the largest value,establish the ACS early warning model and the AUC is calculated.8.Multi-dimensional data were integrated to establish the acute coronary syndrome prediction model using XGBoost method.9.Compare the AUC obtained by different modeling methods and screen the dominant models.Results:1.Comparison of anatomical parameters: there was no significant difference in the baseline values of lesion length,plaque volume and remodeling index between the groups,and the lumen stenosis rate of criminal vessels was higher(55.94 ± 8.97% vs.50.08 ± 12.41%,P =0.034).There was no significant difference in the follow-up values of lesion length and plaque volume.The follow-up values of lumen stenosis rate and remodeling index in ACS group were higher(66.60 ± 11.05% vs.54.75 ± 13.22%,P = 0.001,1.09 ± 0.03 vs.0.97 ± 0.02,P = 0.002).The change of lesion length in ACS group tended to increase,but there was no significant difference(P = 0.051).There was no significant difference in the change of plaque volume between the two groups.The changes of lumen stenosis rate and remodeling index in ACS group were significantly higher(10.18 ± 2.26% vs.3.62 ± 1.41%,P = 0.018,0.15 ± 0.14 vs.0.09± 0.01,P < 0.01).2.Comparison of plaque properties: there was no significant difference in the baseline values of the proportion of fibrous plaque,fibrotic fatty plaque,necrotic core and calcified plaque between the groups.There was no significant difference in the follow-up value of fibrous plaque between the groups.The follow-up value of the proportion of calcified plaque in ACS group decreased,but there was no significant difference(P = 0.06).The proportion of followup values of fibrotic fatty and necrotic core increased significantly in ACS group(22.75 ± 2.85%vs.15.92 ± 1.68%,P = 0.031,13.25 ± 2.23% vs.7.75 ± 1.28%,P = 0.025).There was no significant difference in the changes of fibrous plaque and fibrotic fatty plaque between the groups.The changes of necrotic core and calcified plaque in ACS group were significantly higher and lower(4.79 ± 1.84% vs.0.43 ± 1.09%,P = 0.019,-4.28 ± 2.48% vs.4.48 ± 1.46%,P = 0.004).3.Hemodynamic comparison: there was no significant difference in baseline value between groups,and the follow-up value and change value in ACS group decreased significantly(0.783 ± 0.005 vs.0.823 ± 0.004,P < 0.01,-0.05 ± 0.005 vs.-0.01 ± 0.003,P <0.001).The decrease of CT-FFR in criminal lesions was mainly distributed between 0.04-0.08(63.64%),while the decrease of CT-FFR in non-criminal lesions was mainly distributed between-0.01-0.02(71.42%).4.Using the method of logistic regression,the baseline value,follow-up value and change value of vascular anatomical structure characteristics were included,the early warning model of ACS events was established,and the ROC curve was drawn.The AUC under the curve were0.664(0.593-0.702),0.759(0.711-0.801)and 0.794(0.736-0.827)respectively.5.Using the method of logistic regression,the baseline value,follow-up value and change value of plaque nature were included,the early warning model of ACS events was established,and the ROC curve was drawn.The AUC under the curve were 0.571(0.522-0.634),0.611(0.575-0.668)and 0.721(0.678-0.759)respectively.6.Using the method of logistic regression,the baseline value,follow-up value and change value of hemodynamic CT-FFR were included,the early warning model of ACS events was established,and the ROC curve was drawn.The AUCs under the curve were 0.627(0.574-0.685),0.777(0.723-0.831)and 0.864(0.821-0.917)respectively.7.Using the method of logistic regression,27 parameters(vascular anatomical structure,plaque nature and composition,baseline value of CT-FFR,follow-up data and their change value)were included to establish the early warning model of ACS events,draw the ROC curve,and the area under the curve AUC = 0.890(0.819-0.962).8.Logistic regression was used to screen important variables,among which the variables with the top five or values were: the change value of reconstruction index [OR 4.667(1.712-9.724),P = 0.002],the follow-up value of reconstruction index [OR 4.222(1.541-9.568),P =0.002],the change value of plaque volume [OR 3.000(1.126-7.993),P = 0.026],the change value of CT-FFR [OR2.960(1.270-6.989),P = 0.001],the change value of necrotic core ratio[OR 2.469(1.262-4.827),P = 0.003].Incorporate the above five variables,establish the early warning model of ACS events,draw the ROC curve,and the area under the curve AUC = 0.878(0.803-0.953).9.According to the analysis of shake method,the first five parameters that have the most important impact on the prediction of ACS are: the change value of calcified plaque ratio,the change value of necrotic core ratio,the change value of CT-FFR,the change value of plaque volume and the follow-up value of CT-FFR.The importance scores are 0.143,0.096,0.088,0.080 and 0.054 respectively.Five parameters are included,and the early warning model of ACS events is established by machine learning XGBoost method.The ROC curve is drawn,and the area under the curve AUC = 0.918(0.861-0.968).10.Comparing the calculated AUC,the largest value is XGBoost model,followed by the logistics model with 27 all factors and the logistics model with 5 OR values.Conclusions:1.Compared with the regression modeling method with 27 factors,the prediction model established by XGBoost machine learning with 5 factors can achieve higher AUC.2.Compared with the baseline value and follow-up value of lesion characteristics,the dynamic change value may have higher specificity for early warning of ACS.3.Compared with the plaque nature and vascular structure,the abnormal decrease of hemodynamic changes indicates the high risk of ACS.4.In predicting ACS events,comprehensive analysis based on noninvasive CTA data is better than independent analysis of a single factor. |
PDF Full Text Request