基于多导联心电信号和多变量回归模型的心律失常的分类

心电信号(ECG)智能分析非常有利于严重心脏病人的自动诊断。本文介绍了多变量回归模型(MAR)建模法，利用MAR模型从双导联ECG中提取特征对ECG信号进行分类。在分类时，利用MAR模型系数及其K—L变换(K—LMAR系数)作为信号特征，并采用了树状决策过程和二次判别函数(QDF)分类器。利用文中方法对MIT—BIH标准数据库中的正常窦性心律(NSR)、期收缩(APC)、心室早期收缩(PVC)、心室性心动过速(VT)和心室纤维性颤动(VF)各300个样本信号进行了建模和测试。结果表明，为了达到分类目的，MAR模型阶数取4是足够的，基于MAR系数的分类取得了比基于K—L MAR系数的分类稍好的结果。基于MAR系数的分类获得了97．3％～98．6％的分类精度。

Cardiac arrhythmia classification based on mutiple lead electrocardiogram signals and multivariate autoregressive modeling method

Artificial-intelligence analysis of electrocardiogram (ECG) signals is great benefit to the automatic diagnosis in critical ill patients. Multivariate autoregressive modeling (MAR) for the purpose of classification of cardiac arrhythmias has been introduced. The MAR coefficients and K-L transformation of MAR coefficients extracted from two-lead ECG signals have been utilized for representing the ECG signals. The ECG data obtained from MIT-BIH database included normal sinus rhythm, atria premature contraction, premature ventricular contraction, ventricular tachycardia, and ventricular fibrillation. The current classification was performed using a stage-by-stage quadratic discriminant function (QDF). The results showed a MAR order of 4 was sufficient for the purpose of classification, and MAR coefficients produced slightly better results than K-L transformation of MAR coefficients. The classification accuracy of 97.3% to 98.6% based on MAR coefficients is obtained in the research.