Long short-term memory (LSTM) recurrent neural network for muscle activity detection

<strong>Background: </strong>The accurate temporal analysis of muscle activation is of great interest in many research areas, spanning<br> from neurorobotic systems to the assessment of altered locomotion patterns in orthopedic and neurological<br> patients and the monitoring of their motor rehabilitation. The performance of the existing muscle activity detectors<br> is strongly affected by both the SNR of the surface electromyography (sEMG) signals and the set of features used to<br> detect the activation intervals. This work aims at introducing and validating a powerful approach to detect muscle<br> activation intervals from sEMG signals, based on long short-term memory (LSTM) recurrent neural networks.<br> <strong>Methods: </strong>First, the applicability of the proposed LSTM-based muscle activity detector (LSTM-MAD) is studied<br> through simulated sEMG signals, comparing the LSTM-MAD performance against other two widely used approaches,<br> i.e., the standard approach based on Teager–Kaiser Energy Operator (TKEO) and the traditional approach, used in<br> clinical gait analysis, based on a double-threshold statistical detector (Stat). Second, the effect of the Signal-to-Noise<br> Ratio (SNR) on the performance of the LSTM-MAD is assessed considering simulated signals with nine different SNR<br> values. Finally, the newly introduced approach is validated on real sEMG signals, acquired during both physiological<br> and pathological gait. Electromyography recordings from a total of 20 subjects (8 healthy individuals, 6 orthopedic<br> patients, and 6 neurological patients) were included in the analysis. <strong>Results</strong>: The proposed algorithm overcomes the main limitations of the other tested approaches and it works<br> directly on sEMG signals, without the need for background-noise and SNR estimation (as in Stat). Results demonstrate<br> that LSTM-MAD outperforms the other approaches, revealing higher values of F1-score (F1-score &gt; 0.91) and Jaccard<br> similarity index (Jaccard &gt; 0.85), and lower values of onset/offset bias (average absolute bias &lt; 6 ms), both on simulated<br> and real sEMG signals. Moreover, the advantages of using the LSTM-MAD algorithm are particularly evident for<br> signals featuring a low to medium SNR. <strong>Conclusions</strong>: The presented approach LSTM-MAD revealed excellent performances against TKEO and Stat. The<br> validation carried out both on simulated and real signals, considering normal as well as pathological motor function<br> during locomotion, demonstrated that it can be considered a powerful tool in the accurate and effective recognition/<br> distinction of muscle activity from background noise in sEMG signals.

研究背景：肌肉激活的精准时域分析在诸多研究领域中备受关注，研究范畴涵盖神经机器人系统、骨科与神经科患者异常运动模式评估，以及其运动康复监测。现有肌肉活动检测器的性能，受表面肌电（surface electromyography, sEMG）信号的信噪比（Signal-to-Noise Ratio, SNR）以及用于检测激活区间的特征集双重显著影响。本研究旨在提出并验证一种基于长短期记忆（long short-term memory, LSTM）循环神经网络的sEMG信号肌肉激活区间检测方法，该方法性能优异。
研究方法：首先，通过模拟sEMG信号对所提出的基于LSTM的肌肉活动检测器（LSTM-MAD）的适用性进行研究，并将其性能与另外两种广泛使用的方法进行对比：一是基于泰格-凯撒能量算子（Teager–Kaiser Energy Operator, TKEO）的标准方法，二是临床步态分析中常用的基于双阈值统计检测器（Stat）的传统方法。其次，针对9种不同信噪比水平的模拟sEMG信号，评估信噪比对LSTM-MAD性能的影响。最后，在生理步态与病理步态下采集的真实sEMG信号上，对该新提出的方法进行验证。本分析共纳入20名受试者的肌电记录数据，其中健康志愿者8名、骨科患者6名、神经科患者6名。
研究结果：所提出的算法克服了其他受试方法的主要局限，可直接对sEMG信号进行处理，无需如Stat方法那般进行背景噪声与信噪比估计。结果表明，无论在模拟还是真实sEMG信号上，LSTM-MAD的性能均优于其他方法：其F1分数（F1-score > 0.91）与雅卡尔相似性指数（Jaccard similarity index > 0.85）均更高，而起始/偏移偏差的平均绝对偏差更低（<6 ms）。此外，LSTM-MAD算法的优势在低至中等信噪比的信号上尤为显著。
研究结论：相较于TKEO与Stat方法，本文提出的LSTM-MAD方法展现出优异的性能。本研究通过模拟信号与真实信号（涵盖正常运动与病理运动步态场景）开展的验证实验表明，该方法可作为一种高效工具，精准且有效地从sEMG信号中识别并区分肌肉活动与背景噪声。