目的 探讨基于心脏磁共振(cardiovascular magnetic resonance,CMR)不同区域心肌的影像组学特征及相关临床因素在预测肥厚型心肌病(hypertrophic cardiomyopathy,HCM)患者并发室性心律失常(ventricular arrhythmias,VAs)中的价值.材料与方法 回顾性分析宁夏医科大学总医院2018年1月1日至2023年5月31日的122例确诊HCM患者的CMR图像及临床资料.根据24 h动态心电图(24-hour dynamic electrocardiogram,24 h DCG)结果将其分为合并VAs组(40例)与未合并VAs组(82例).按照7∶3的比例将所有受试者分为训练集与测试集,训练集用于构建模型,测试集用于评估模型效能.选择从二尖瓣水平至心尖部的左心室短轴未增强亮血电影序列,以舒张末期壁厚为基础,将HCM患者心肌分为肥厚区域与非肥厚区域.在左心室短轴各层面心肌的心室舒张末期进行全室壁心肌、肥厚区域心肌及非肥厚区域心肌的感兴趣区勾画及影像组学特征提取.采用Mann-Whitney U检验、递归特征消除法及最小绝对收缩与选择算法进行影像组学特征筛选、构建影像组学模型、建立影像组学标签并计算影像组学评分;通过logistic回归分析筛选临床危险因素建立临床因素模型;并构建基于临床危险因素和影像组学特征的联合模型.采用准确度、敏感度、特异度、阳性预测值、阴性预测值和受试者工作特征曲线下面积(area under the curve,AUC)及DeLong检验评估和比较模型间的预测效能;使用列线图将联合模型可视化,并通过拟合优度Hosmer-Lemeshow检验及校正曲线评估联合模型拟合程度;通过决策曲线分析观察联合模型的临床实用性.结果 在训练集中,左心室非肥厚区域心肌联合模型(AUC值=0.89)、左心室肥厚区域联合模型(AUC值=0.98)、左心室全心肌联合模型(AUC值=0.98)的AUC值高于左心室非肥厚区域心肌影像组学模型(AUC值=0.74)、左心室肥厚区域影像组学模型(AUC值= 0.85)、左心室全心肌影像组学模型(AUC值=0.86)(P<0.05),亦高于临床因素模型(AUC值=0.81)(P<0.05);左心室非肥厚区域心肌联合模型(AUC值=0.89)的AUC值小于左心室肥厚区域心肌联合模型(AUC值=0.98)及左心室全心肌联合模型(AUC值=0.98)(P<0.05),左心室肥厚区域心肌联合模型与左心室全心肌联合模型的AUC值差异无统计学意义(P>0.05).在测试集中,左心室非肥厚区域心肌影像组学模型(AUC值=0.75)的AUC值低于左心室全心肌联合模型(AUC值=0.93)及左心室肥厚区域心肌联合模型(AUC值=0.95)(P<0.05);左心室肥厚区域心肌联合模型(AUC值=0.95)的AUC值高于左心室非肥厚区域心肌联合模型(AUC值=0.80)(P<0.05).结论 基于CMR不同区域心肌的影像组学特征及相关临床因素所构建的模型在预测HCM者合并VAs风险方面有重要价值.
Value of CMR radiomics combined with clinical factors in predicting hypertrophic cardiomyopathy complicated by ventricular arrhythmias
Objective:To explore the value of radiomics features of myocardium in different regions based on cardiac magnetic resonance(CMR)and related clinical factors in predicting hypertrophic cardiomyopathy complicated by ventricular arrhythmias.Materials and Methods:The CMR images and clinical data of 122 patients diagnosed with HCM from January 1,2018 to May 31,2023 were retrospectively collected and analyzed.According to the results of 24-hour dynamic electrocardiogram(24 h DCG),the patients were divided into a VAs combined group(40 cases)and a VAs uncombined group(82 cases).All subjects were divided into a training set and a test set according to a ratio of 7∶3.The training set was used to build the model,and the test set was used to evaluate the model efficacy.The left ventricular short axis unenhanced bright blood cine sequence from mitral valve level to apex was selected.On the basis of end-diastolic wall thickness,the myocardium of HCM patients was divided into hypertrophic regions and non-hypertrophic regions.At the end of ventricular diastole in all layers of left ventricular short axis myocardium,the delineation of areas of interest and the extraction of radiomics features of the whole ventricular wall myocardium,the myocardium in hypertrophic regions and non-hypertrophic regions were performed.Mann-Whitney U test,recursive feature elimination method and least absolute shrinkage and selection operator were used to screen the radiomics features,construct the radiomics model,establish the radiomics label and calculate the radiomics score.The clinical risk factors were screened by logistic regression analysis to establish the clinical factor model,and combined model which based on clinical risk factors and radiomcs features was constructed.Accuracy,sensitivity,specificity,positive predictive value,negative predictive value,area under the curve(AUC)and DeLong test were used to evaluate and compare the predictive power of models.The combined model was visualized by using a nomogram,and the fitting degree of the combined model was evaluated by Hosmer-Lemeshow test(H-L test)and calibration curve.The clinical practicability of the combined model was observed through decision curve analysis.Results:In the training set,the AUC values of the combined model of the left ventricular non-hypertrophic region(AUC=0.89),combined model of the left ventricular hypertrophic region(AUC=0.98),combined model of the left ventricular whole myocardium(AUC=0.98)were higher than those of the radiomics model of left ventricular non-hypertrophic regions myocardium(AUC=0.74)and the radiomics model of left ventricular hypertrophic regions myocardium(AUC=0.85)and the radiomics model of left ventricular whole myocardium(AUC=0.86)(P<0.05)were higher than clinical factor model(AUC value=0.81)(P<0.05).The AUC value of the combined model of left ventricular non-hypertrophic regions myocardium(AUC=0.89)was lower than that of the combined model of left ventricular hypertrophic regions myocardium(AUC=0.98)and the combined model of left ventricular whole myocardium(AUC=0.98)(P<0.05).There was no significant difference in the AUC value between the combined model of left ventricular hypertrophy and the combined model of left ventricular whole myocardium(P>0.05).In the test set,the AUC value of the radiomics model of left ventricular non-hypertrophic regions myocardium(AUC=0.75)was lower than that of the combined model of left ventricular whole myocardium(AUC=0.93)and the combined model of left ventricular hypertrophic regions myocardium(AUC=0.95)(P<0.05).The AUC of the combined model of left ventricular hypertrophic regions myocardium(AUC=0.95)was higher than that of the combined model of left ventricular non-hypertrophy regions myocardium(AUC=0.80)(P<0.05).Conclusions:The model based on the radiomics features of myocardium in different regions of CMR and related clinical factors has important value in predicting the risk of HCM patients complicated by VAs.
hypertrophic cardiomyopathyventricular arrhythmiasprediction modelcardiac magnetic resonanceradiomicsmagnetic resonance imaging
国家科技期刊平台
NSTL
万方数据
维普
