Romero Inaki, Grundlehner Bernard, Penders Julien
Holst Centre / IMEC-NL, High Tech Campus 31, Eindhoven, Netherlands.
Annu Int Conf IEEE Eng Med Biol Soc. 2009;2009:950-3. doi: 10.1109/IEMBS.2009.5334543.
Robust beat detection under noisy conditions is required in order to obtain a correct clinical interpretation of the ECG in ambulatory settings. This paper describes the evaluation and optimization of a beat detection algorithm that is robust against high levels of noise. An evaluation protocol is defined in order to study four different characteristics of the algorithm: non-rhythmic patterns, different levels of SNR, exact peak detection and different levels of physical activity. This protocol is based on the MIT/BIH arrhythmia database and additional ECG recordings obtained under different levels of physical activity measured by 2-axis accelerometers. The optimized algorithm obtained a Se=99.65% and +P=99.79% on the MIT/BIH arrhythmia database while keeping a good performance on ECGs with high levels of activity (overall of Se=99.86%, +P=99.91%). In addition, this method was optimized to work in real time, for future implementation in a Wireless ECG sensor based on a microprocessor.
为了在动态环境中对心电图进行正确的临床解读，需要在噪声环境下进行稳健的心跳检测。本文描述了一种对高水平噪声具有鲁棒性的心跳检测算法的评估和优化。定义了一个评估协议，以研究该算法的四个不同特性：非节律模式、不同的信噪比水平、精确的峰值检测和不同的身体活动水平。该协议基于麻省理工学院/贝斯以色列女执事医疗中心心律失常数据库以及通过双轴加速度计在不同身体活动水平下获得的额外心电图记录。优化后的算法在麻省理工学院/贝斯以色列女执事医疗中心心律失常数据库上的灵敏度（Se）为99.65%，阳性预测值（+P）为99.79%，同时在高活动水平的心电图上保持良好性能（总体灵敏度为99.86%，阳性预测值为99.91%）。此外，该方法经过优化可实时运行，以便未来在基于微处理器的无线心电图传感器中实现。
