Explore The Value Of The Predictive Model Constructed By Computed Tomography Angiography Radiomics Features In Identifying Symptomatic Carotid Plaques

Objective:To evaluate and verify the value of predictive model based on radiomics features of non-invasive head and neck computed tomography angiography(CTA)in identifying symptomatic carotid plaques,and to explore the potential of radiomics techniques in risk stratification of carotid plaques.Methods:A total of 191 patients with extracranial carotid atherosclerosis who underwent head and neck CTA examination in the Imaging Department of the first Hospital of Shanxi Medical University from January 2018 to December 2021 were analyzed retrospectively.Extracted the traditional quantitative features and radiomics features of carotid plaques were from CTA images.The optimal radiomics features were screened by intra-class correlation coefficient(ICC),spearman rank correlation and least absolute shrinkage and selection operator regresion analysis(LASSO),and the radiomics score(Radscore)was calculated.Patients were divided into symptomatic group and asymptomatic group according to whether they had ischemic cerebrovascular events recently.The clinical data of the two groups were compared,and the traditional quantitative features independently related to symptomatic carotid plaque were screened by univariate and multivariate analysis.According to a proportion of 7:3,patients were randomly divided into training set(n = 134)and test set(n = 57).The traditional risk model and rad-score model were constructed by R software,Discriminatory value of models were determined using receiver operating characteristic(ROC)analysis,area under the curve(AUC).The calibration degree of rad-score model was evaluated by calibration curve(Hosmer-Lemeshow good of fit test).Results:Among the 191 patients,123 patients had cerebrovascular symptoms and 68 patients had were asymptomatic.Compared with the asymptomatic group,patients with hyperhomocysteinemia were more in symptomatic group(p<0.05),Carotid atherosclerotic patients with hyperhomocysteinemia have a significantly increased risk of cerebrovascular symptoms(OR,4.98;95%CI,2.51～9.90,p<0.05).In multivariate analysis of traditional quantitative features,ulceration(OR=4.35,95%CI:1.62～11.71,p<0.05),plaque thickness(OR=1.67,95%CI:1.08～2.58,p<0.05)and degree of luminal stenosis(OR=1.03,95%CI:1.01～1.05,p<0.05)were independently related to symptomatic carotid plaque,which were used to construct traditional risk model.Finally,eleven radiomics features were used to calculate rad-score and construct rad-score model.The AUC of traditional risk model and rad-score model in training set and test set were medium to good(AUC=0.72,0.71 and 0.91,0.89,respectively),the effectiveness of Radscore model was significantly better than that of traditional risk model(p < 0.05),and the calibration of rad-score model in training set and test set was good(p> 0.05).Conculsion:Clinical attention should be paid to carotid atherosclerosis patients with hyperhomocysteinemia and to guard against the occurrence of ischemic cerebrovascular events.The predictive model derived from CTA radiomics features can effectively discriminate symptomatic and asymptomatic carotid plaques,which was significantly better than the current quantitative features.Radiomics techniques based on CTA might serve as an important non-invasive tool for improving risk stratification of patients with carotid plaques.