基于瞬时频率与功率谱熵组合特征的心律失常诊断方法

An Arrhythmia Diagnosis Method Based on the Combined Feature of Instantaneous Frequency and Power Spectrum Entropy

心率失常作为最常见的心血管疾病,威胁着广大人民群众的身体健康。近年来,通过机器学习对心律失常进行诊断逐渐成为热点,然而,当前方法很难在准确率和效率之间取得平衡。为解决这一问题,本文从实用角度出发,提出了一种基于瞬时频率与功率谱熵组合特征的心律失常诊断方法,并使用MIT-BIH心律失常数据库中的心电图数据对该方法进行测试。本文采用Pan-Tompkins算法定位心拍位置以构成数据集,并利用小波分解对数据进行降噪,之后将瞬时频率与功率谱熵作为特征,使用支持向量机(SVM)与双向长短期记忆网络(Bi-LSTM)以测试其性能。最终可得该方法在SVM和Bi-LSTM下的准确率分别为98.3%和99%,识别速度均在毫秒级,同时兼顾了效率与正确率,为应用机器学习的心律失常诊断提供了一种更加优秀的解决方案。

As the most common cardiovascular disease, arrhythmia is a threat to the health of the public. In recent years, the diagnosis of arrhythmia through machine learning has gradually become a hot spot, however, current methods are difficult to strike a balance between accuracy and efficiency. To solve this problem, an arrhythmia diagnosis model based on instantaneous frequency and power spectrum entropy is proposed by writers, and the ECG data from MIT-BIH arrhythmia database is used to test this model. In this research, the Pan-Tompkins algorithm was used to locate the cardiac beat to form the data set, and wavelet decomposition was used to denoise the data. Then, the instantaneous frequency and power spectrum entropy were combined as features, and the Support Vector Machine (SVM) and Bidirectional Long Short Term Memory network (Bi-LSTM) were used to test the model performance. Finally, the accuracy of this model respectively is 98.3% and 99% under SVM and Bi-LSTM, and the recognition speed is in millisecond level. Both efficiency and accuracy are good in this method, which provides a more excellent solution for arrhythmia diagnosis using machine learning.