基于多尺度小波熵的阵发性房颤识别方法

Multi-scale Wavelet Entropy Based Method for Paroxysmal Atrial Fibrillation Recognition

目的通过多尺度小波熵对阵发性房颤心率变异信号进行量化评估分析,提取特征参数,完成其与远离阵发性房颤信号的识别分类。方法获取心率变异信号和其对应的小波分解系数,筛选出在不同类别间存在显著差异的多尺度小波熵值,利用支持向量机完成识别分类。结果对MIT-BIT PAF Prediction Database 50例表面心电信号进行验证。本文方法与时域、频域以及非线性方法相比,正确率、特异性以及敏感性都显著提升。结论多尺度小波熵能够作为反映阵发性房颤发生过程中心脏内交感神经和迷走神经兴奋活动变化的动态指标,为航天医学人工智能心电监护以及房颤的临床治疗提供参考。

Objective To identify and distinguish the paroxysmal atrial fibrillation （PAF） ECG signal and the signal distant from PAF by using multi-scale wavelet entropy via extracting the characteristic parameters from heart rate variability signals. Methods HRV signals extracted from ECG signals were decomposed into a series of wavelet parameters, which were calculated to get multi-scale wavelet entropies. These entropies were further selected through statistical t-test method with significant difference. Recognition and prediction of PAF signals were achieved by SVM. Results Fifty signals from MIT-BIT PAF prediction database were chosen to verify the proposed algorithm. The correct rate, specificity and sensitivity were greatly improved as compared with the existed studies based on timedomain, frequency-domain or non-linear methods. Conclusion The multi-scale wavelet entropy was quite suitable to be a dynamic indicator for manifesting the active variation of sympathetic nerve and vagus nerve during atrial fibrillation, and it could become a meritorious diagnosable reference for aerospace medicine in terms of artificial intelligent monitoring and clinical treatments.