基于RR间期构建常见心律失常Informer分类模型

    Constructing an Informer classification model for common arrhythmias based on RR intervals

      摘要
    • 摘要:
      目的 基于RR间期数据，构建心房颤动（房颤）与期前收缩（早搏）的Informer分类模型，并进行评价。
      方法 收集并筛选窦性心律、持续性房颤、频发房性早搏、频发室性早搏的动态心电图共150例为研究对象，提取RR间期序列，并截取为50拍的片段，随机选取得到训练集（n=8941）、验证集（n=2235）、测试集（n=3013），构建Informer、VGG16分类模型，并进行分类性能评价。
      结果 与SR组比较，AF组平均心率升高（P<0.01）；与AF组比较，PAC组和PVC组平均心率降低（均P<0.01）； Informer模型分类总正确率为91.04%，对AF的敏感性、阴性预测值最高，分别为99.77%、99.90%，对SR的特异性、阳性预测值最高，分别为99.47%、98.74%，对SR的整体诊断效能最优，准确率为96.81%。Informer与VGG16对SR、PVC的分类性能没有差异，对PACC（P<0.01）的分类性能优于VGG16，对AF（P<0.05）的分类性能不及VGG16。Informer模型与心电专家的诊断结果之间具有高度一致性（Kappa=0.8784）。
      结论 基于RR间期时序性数据构建的Informer模型能够实现房颤、早搏等常见心律失常的分类。

       

      Abstract:
      AIM  Based on RR interval data, an Informer classification model for atrial fibrillation (AF) and premature contractions (premature beats) was constructed and evaluated.
      METHODS A total of dynamic electrocardiograms of 150 patients with sinus rhythm, persistent atrial fibrillation, frequent atrial premature beats, and frequent ventricular premature beats were collected. RR interval sequences were extracted and 50 beat segments were cut. The training set (n=8941), validation set (n=2235), and test set (n=3013) were randomly selected to construct Informer and VGG16classification models, and their classification performance was evaluated.
      RESULTS Compared with the SR group, the AF group showed an increase in average heart rate (P<0.01); Compared with the AF group, the average heart rate of the PAC group and PVC group decreased (both P<0.01); The overall classification accuracy of the Informer model is 91.04%, with the highest sensitivity and negative predictive value for AF at 99.77% and 99.90%, respectively. The specificity and positive predictive value for SR are the highest at 99.47% and 98.74%, respectively. The overall diagnostic performance for SR is the best, with an accuracy of 96.81%. There is no difference in the classification performance of Informer and VGG16 for SR and PVC, but the classification performance of informer for PAC is better than VGG16 (P<0.01), and the classification performance of informer for AF is inferior to VGG16 (P<0.05). There is a high degree of consistency between the Informer model and the diagnostic results of electrocardiogram experts (Kappa=0.8784).
      CONCLUSION The Informer model constructed based on RR interval temporal data can classify common arrhythmias such as atrial fibrillation and premature beats.

       

      • HTML全文
      • 参考文献(21)
      • 相关文章
      • 施引文献
    • 资源附件(0)

    /

    返回文章
    返回