Xiong Jinle, Liang Xueyu, Zhao Lina, Lo Benny, Li Jianqing, Liu Chengyu
School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China.
The Hamlyn Centre/Department Surgery and Cancer, Imperial College London, London SW7 2AZ, UK.
Entropy (Basel). 2020 May 2;22(5):520. doi: 10.3390/e22050520.
Due to the wide inter- and intra-individual variability, short-term heart rate variability (HRV) analysis (usually 5 min) might lead to inaccuracy in detecting heart failure. Therefore, RR interval segmentation, which can reflect the individual heart condition, has been a key research challenge for accurate detection of heart failure. Previous studies mainly focus on analyzing the entire 24-h ECG recordings from all individuals in the database which often led to poor detection rate. In this study, we propose a set of data refinement procedures, which can automatically extract heart failure segments and yield better detection of heart failure. The procedures roughly contain three steps: (1) select fast heart rate sequences, (2) apply dynamic time warping (DTW) measure to filter out dissimilar segments, and (3) pick out individuals with large numbers of segments preserved. A physical threshold-based Sample Entropy (SampEn) was applied to distinguish congestive heart failure (CHF) subjects from normal sinus rhythm (NSR) ones, and results using the traditional threshold were also discussed. Experiment on the PhysioNet/MIT RR Interval Databases showed that in SampEn analysis (embedding dimension = 1, tolerance threshold = 12 ms and time series length = 300), the accuracy value after data refinement has increased to 90.46% from 75.07%. Meanwhile, for the proposed procedures, the area under receiver operating characteristic curve (AUC) value has reached 95.73%, which outperforms the original method (i.e., without applying the proposed data refinement procedures) with AUC of 76.83%. The results have shown that our proposed data refinement procedures can significantly improve the accuracy in heart failure detection.
由于个体间和个体内存在广泛的变异性，短期心率变异性（HRV）分析（通常为5分钟）可能导致心力衰竭检测不准确。因此，能够反映个体心脏状况的RR间期分割，一直是准确检测心力衰竭的关键研究挑战。以往的研究主要集中在分析数据库中所有个体的完整24小时心电图记录，这往往导致检测率较低。在本研究中，我们提出了一套数据细化程序，该程序可以自动提取心力衰竭片段，并能更好地检测心力衰竭。这些程序大致包括三个步骤：（1）选择快速心率序列，（2）应用动态时间规整（DTW）方法滤除不相似的片段，（3）挑选保留片段数量较多的个体。应用基于物理阈值的样本熵（SampEn）来区分充血性心力衰竭（CHF）患者和正常窦性心律（NSR）患者，并讨论了使用传统阈值的结果。在PhysioNet/MIT RR间期数据库上的实验表明，在SampEn分析中（嵌入维数 = 1，容差阈值 = 12毫秒，时间序列长度 = 300），数据细化后的准确率从75.07%提高到了90.46%。同时，对于所提出的程序，接收器操作特征曲线（AUC）值达到了95.73%，优于原始方法（即未应用所提出的数据细化程序）的AUC值76.83%。结果表明，我们提出的数据细化程序可以显著提高心力衰竭检测的准确性。
