基于超声影像组学和临床特征的列线图预测心肌梗死后左室壁瘤患者预后的临床价值

Clinical value of a nomogram based on ultrasonic radiomics and clinical characteristics for predicting the prognosis of patients with left ventricular aneurysm after myocardial infarction

目的基于超声影像组学特征和临床特征构建列线图模型,探讨其预测急性心肌梗死后左室壁瘤(PILVA)患者远期预后的临床应用价值。方法选取在我院心血管中心住院治疗的PI-LVA患者153例,根据随访24、36、60个月是否出现终点事件(心源性死亡)分为事件组38例和非事件组115例,比较两组一般资料和超声检查结果的差异。采用多因素Cox回归分析筛选与PI-LVA患者远期预后相关的影响因素。采集患者心尖四腔心切面收缩期且包含完整室壁瘤的图像,使用ITK-SNAP软件进行分割并获取感兴趣区,Pyradiomics软件包提取超声影像组学特征,并采用最小绝对收缩和选择算子(LASSO)回归筛选对诊断有利的特征子集及对应非零系数。基于超声影像组学特征及多因素Cox回归分析筛选的影响因素构建预测模型并绘制列线图,绘制受试者工作特征(ROC)曲线、校准曲线、决策曲线分别评估列线图模型预测PI-LVA患者随访24、36、60个月出现终点事件的区分度、校准度及临床适用性。结果事件组与非事件组左室射血分数(LVEF)<35%、右冠状动脉重度狭窄占比及Tp-Te间期比较差异均有统计学意义(均P<0.05)。多因素Cox回归分析显示,LVEF<35%、右冠状动脉重度狭窄、Tp-Te间期均为与PI-LVA患者远期预后相关的影响因素(均P<0.05)。共提取107个超声影像组学特征,经LASSO回归分析筛选出3个显著性特征。基于LVEF<35%、右冠状动脉重度狭窄、Tp-Te间期及超声影像组学特征构建预测模型,并绘制列线图可视化。ROC曲线分析显示列线图模型预测PI-LVA患者随访24、36、60个月出现终点事件的曲线下面积分别为0.850、0.831、0.898;校准曲线显示列线图模型预测概率与实际概率的一致性良好;决策曲线显示列线图模型具有较好的临床适用性。结论基于超声影像组学特征和临床特征构建的列线图模型可有效预测PI-LVA患者远期预后,具有较好的临床应用价值。

Objective To construct a nomogram model based on ultrasonic radiomics and clinical characteristics,and to explore its clinical application value in predicting the long-term prognosis of patients with left ventricular aneurysm after acute myocardial infarction(PI-LVA).Methods A total of 153 patients with PI-LVA from the cardiovascular centre of Suining Central Hospital were selected.They were divided into event group(38 cases)and non-event group(115 cases)according to the followed up with the endpoint event of cardiac death at 24th,36th,60th months,and the differences in general data and ultrasonographic results were compared.Multifactorial Cox regression analysis was used to screen the influencing factors related to the long-term prognosis of patients with PI-LVA.Images of ventricular wall tumour from the apical of four-chamber cardiac sections were captured,regions of interest were segmented and acquired by ITK-SNAP software,and ultrasonic radiomics features were extracted by Pyradiomics,the feature subset and corresponding non-zero coefficients were screened by least absolute systolic and selection operator(LASSO)regression.The prediction model was constructed and plotted as a nomogram based on the ultrasonic radiomics and influencing factors screened by multifactorial Cox regression analysis,and the differentiation of the model for predicting the endpoint event at 24th,36th,60th months in patients with PI-LVA was analyzed by receiver operating characteristic(ROC)curve.The calibration curve and decision curve were applied to assess the calibration degree and clinical applicability of the model.Results There were statistical significance in proportion of left ventricular ejection fraction(LVEF)<35%,severe right coronary artery stenosis and Tp-Te interval between event and non-event groups(all P<0.05).Multifactorial Cox regression analysis showed that LVEF<35%,severe right coronary artery stenosis and Tp-Te interval were independent influencing factors on the long-term prognosis of patients with PI-LVA.A total of 107 ultrasonic radiomics features were extracted,three significant features were screened and selected by LASSO regression analysis.A prediction model was constructed based on LVEF<35%,severe right coronary artery stenosis,Tp-Te interval and ultrasonic radiomics features,then a nomogram was plotted for visualisation.ROC curve analysis showed that the area under the curve of nomogram model for predicting the endpoint event at 24th,36th,60th months in patients with PI-LVA were 0.850,0.831 and 0.898,respectively.Calibration curve showed that the predicted probability was consistent with the actual probability,and decision curve analysis showed that the nomogram had good clinical applicability.Conclusion The nomogram model based on ultrasonic radiomics and clinical characteristics can effectively predict the long-term prognosis of patients with PI-LVA,and has good clinical application value.