基于CCTA的冠状动脉周围脂肪组织影像组学特征预测急性冠状动脉综合征

Using CCTA-based pericoronary adipose tissue radiomics to predict acute coronary syndrome

目的评估基于冠状动脉CT血管成像(CCTA)的冠状动脉周围脂肪组织(PCAT)影像组学特征对疑似冠心病病人2年内发生急性冠状动脉综合征(ACS)的预测能力。方法回顾性收集接受CCTA检查的疑似冠心病病人,将CCTA检查后2年内发生ACS的病人作为ACS组(81例),2年内未发生ACS的疑似冠心病病人作为对照组(81例)。ACS组年龄44~85岁,平均(64.01±10.09)岁,男57例;对照组年龄39~89岁,平均(62.91±10.11)岁,男56例。将2组病人随机以3∶1的比例分为训练集(ACS组60例,对照组60例)和验证集(ACS组21例,对照组21例)。筛选基于CCTA的PCAT影像组学特征,采用多因素Logistic回归分析构建PCAT影像组学评分模型,并基于PCAT影像组学特征中的CT密度值建立PCAT密度模型。采用DeLong检验比较不同数据集中2个模型的诊断效能差异。采用受试者操作特征(ROC)曲线、校准曲线及决策曲线比较2种模型的预测效能。结果训练集和验证集中,ACS组和对照组病人的临床资料间差异均无统计学意义(均P>0.05)。从基于CCTA影像所示的冠状动脉斑块周围PCAT共提取107个影像组学特征,最终筛选出21个最优影像组学特征,包括形态学特征5个、直方图特征1个、纹理特征15个,采用Logistic回归分析构建PCAT影像组学评分模型。基于提取的PCAT组学特征中平均CT密度值构建PCAT密度模型。2种模型预测2年内发生ACS事件的诊断效能分析显示,PCAT影像组学评分模型在训练集及验证集中的曲线下面积(AUC)(AUC=0.841,0.839)均高于PCAT密度的AUC(AUC=0.603,0.588)。训练集中,PCAT影像组学评分的诊断效能优于PCAT密度模型(P<0.05),并在验证集中得到验证(P<0.05)。PCAT影像组学评分对发生ACS事件的预测结果与实际结果一致性高于PCAT密度。PCAT影像组学评分的临床应用价值显著优于PCAT密度。结论基于CCTA的PCAT影像组学特征可为ACS事件的发生提供更多的预测信息。PCAT影像组学评分对2年内发生ACS事件的预测能力显著优于PCAT密度。

Objective To evaluate the ability of CCTA-based pericoronary adipose tissue(PCAT)to predict the occurrence of acute coronary syndrome(ACS)within 2 years in patients with suspected coronary artery disease.Methods The patients with suspected coronary artery disease who underwent CCTA scanning were retrospectively collected.There were 81 patients with an ACS event within 2 years were assigned as ACS group,and 81 patients without ACS during 2 years follow-up were assigned as control group.In the ACS group,there were 57 males,aged from 44 to 85 years old,with an average age of 64.01±10.09 years.In the control group,there were 56 males,aged from 39 to 89 years old,with an average age of 62.91±10.11 years.Two groups were randomly divided into training(ACS=60,control=60)and testing dataset(ACS=21,control=21)with a ratio of 3∶1.We extracted and selected radiomics features from PCAT surrounding plaques based on CCTA images,and construct a PCAT radiomics score model through multiple Logistic regression analysis.CT attenuation derived from PCAT radiomics features was used to build PCAT attenuation model.DeLong test was used to compare the diagnostic performance of two models in different datasets.Receiver operating characteristic(ROC),calibration and decision curves were applied to compare the predictive performance of two models.Results Regardless of the training or testing dataset,there was no statistically significant difference in clinical data between the ACS and control group,and between the training and testing dataset(all P>0.05).A total of 107 radiomics features were extracted from PCAT around coronary artery plaques based on CCTA images,and 21 radiomics features were finally selected,including 5 shape features,1 histogram feature,and 15 texture features,which were used to construct PCAT radiomics score by multivariate Logistic regression analysis.PCAT attenuation was constructed based on the average CT attenuation derived from the extracted PCAT radiomics features.Regarding the diagnostic efficacy of two models in predicting ACS within 2 years,ROC curves showed that the AUCs of PCAT radiomics score in the training and testing dataset(AUC=0.841,0.839)were higher than PCAT attenuation(AUC=0.603,0.588).In the training dataset,the diagnostic performance of PCAT radiomics score was better than that of PCAT attenuation(P<0.05),which was then confirmed in the testing dataset(P<0.05).PCAT radiomics score had a good fitness between prediction and observation for the probability of ACS in both training and test datasets.PCAT radiomics score provided more clinical benefit than PCAT attenuation.Conclusion PCAT radiomics score based on CCTA could provide more predictive information for the occurrence of ACS,PCAT radiomics score significantly outperforms PCAT attenuation in predicting the occurrence of ACS within future 2 years.