文献速递心血管MR

基于机器学习模型预测肥厚型心肌病患者主要不良心脏事件

Machine Learning-based Model for Major Adverse Cardiac Event Prediction in Patients with Hypertrophic Cardiomyopathy.
PubMed 文献 · Radiol Cardiothorac Imaging · 2026-01-01

本研究回顾性纳入604例接受心脏MRI评估的肥厚型心肌病患者,开发基于弹性网络惩罚Cox回归的机器学习模型,用于预测主要不良心脏事件。模型C指数为0.75,与欧洲心脏病学会模型表现相似。心脏MRI特征在危险分层中具有重要作用。

要点

  • 心尖室壁瘤、广泛LGE、左室收缩末期容积指数是MACE关键影像预测因子
  • 心脏MRI常规序列及native T1 mapping可有效辅助HCM危险分层
  • 男性为MACE的独立临床危险因素,报告中应标注

核心结论

  • 基于常规可获得的临床、遗传、超声心动图及心脏MRI变量构建的机器学习模型对HCM患者MACE预测表现良好,C指数为0.75
  • 心脏MRI特征在HCM危险分层中具有重要影响,关键变量包括心尖室壁瘤、左室收缩末期容积指数、广泛LGE及native T1 z分数

研究设计与样本

  • 心脏MRI包括平衡式稳态自由进动电影序列、native T1和T2 mapping及钆延迟增强
  • MACE定义为心血管死亡、心脏骤停复苏成功或心力衰竭住院的复合终点
  • 采用弹性网络惩罚Cox比例风险模型,纳入33个临床、遗传、超声心动图及心脏MRI变量
  • 模型训练使用200次分层子抽样交叉验证,80%训练、20%测试

关键影像表现/征象

  • 心尖室壁瘤为MACE预测的关键影像征象之一
  • 广泛钆延迟增强为MACE预测的重要影像表现
  • 左室收缩末期容积指数体表面积校正值为MACE相关的心脏MRI定量参数
  • native T1 z分数为MACE预测的关键心肌组织特征参数

诊断标准与关键数值

  • CoxNet模型预测MACE的C指数为0.75,95%CI为0.65至0.83
  • 欧洲心脏病学会模型预测MACE的C指数为0.67,95%CI为0.57至0.75
  • 中位随访时间为3.0年,四分位间距为1.9至4.2年

临床应用与阅片要点

  • 阅片时应关注心尖室壁瘤及广泛LGE的存在,两者为MACE关键预测征象
  • 报告应提供左室收缩末期容积指数和native T1 z分数以辅助危险分层
  • 男性HCM患者应视为MACE较高风险群体
  • 该模型表现与现有欧洲心脏病学会风险模型相似,可辅助临床决策

局限性

  • 单中心回顾性设计
  • 模型性能与既有风险模型比较差异无统计学意义,临床增益需进一步验证

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

Purpose To develop a machine learning-based model to identify patients with hypertrophic cardiomyopathy (HCM) at high risk of major adverse cardiac events (MACEs) and key predictors of model performance. Materials and Methods This retrospective cohort study included patients who underwent cardiac MRI for HCM evaluation between September 2015 and December 2022. Cardiac MRI included balanced cine steady-state free precession, native T1 and T2 mapping, and late gadolinium enhancement. MACEs were defined as a composite of cardiovascular death, resuscitated sudden cardiac death, or heart failure hospitalization. A penalized Cox proportional hazards model with elastic net regularization (CoxNet) was developed with 33 clinical, genetic, echocardiography, and cardiac MRI variables. Model training used 200 iterations of stratified subsampling cross-validation (80% training, 20% testing). Performance was evaluated with the Harrell C index and compared with the 2014 European Society of Cardiology sudden cardiac death risk model. Results A total of 604 patients were included (mean age, 52 years ± 15 [SD]; 417 male patients; median follow-up, 3.0 years [IQR, 1.9-4.2 years]). The CoxNet model demonstrated favorable performance for MACE prediction (C index, 0.75; 95% CI: 0.65, 0.83), similar to the European Society of Cardiology model (C index, 0.67; 95% CI: 0.57, 0.75; P = .07). Key predictors included apical aneurysm, left ventricular end-systolic volume indexed to body surface area, extensive late gadolinium enhancement, native T1 z score, and male sex. Conclusion A machine learning-based model comprising routinely available variables showed strong performance for MACE prediction in HCM. Key variables highlight the impact of cardiac MRI features on risk stratification. Keywords: MRI, Machine Learning, Model Training, Cardiac, Heart, Hyperplasia, Hypertrophy © RSNA, 2026 See also commentary by Shiwani in this issue.

来源:Radiol Cardiothorac Imaging,查看原文。版权归原作者所有。

引用

  1. Geyer, McIntosh, Sood, et al. Machine Learning-based Model for Major Adverse Cardiac Event Prediction in Patients with Hypertrophic Cardiomyopathy. Radiol Cardiothorac Imaging. 2026. PMID: 42240460. doi:10.1148/ryct.250433

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。