An Attention-based Deep CNN-BiLSTM Model for Forecasting of Fatigue-induced Surface Electromyography Signals During Isotonic Contractions

Abstract

<p> An accurate estimation of muscle fatigue is critical for adaptive control of existing assistive devices, such as an exoskeleton, prosthesis, and functional electrical stimulation (FES)-based neuroprostheses. However, the estimation of muscle fatigue using surface electromyography (sEMG) for a long duration of time becomes challenging due to loosening of sEMG sensors, sweating, and other accidental failures. These problems can be potentially solved by forecasting future sEMG signals using initially recorded high-quality data points. For the first time, we attempt to forecast the fatigue-induced electromyography signal using the initial sEMG recorded for a shorter interval of time, during biceps curl with weights of 1 kg, 2 kg, 3 kg, and 4 kg. An attention-based deep CNN-BiLSTM neural network model that captures input sEMG dynamics to forecast future sEMG signals corresponding to fatigue state was trained and tested. An average mean absolute percentage error (MAPE) of 26.7% between forecasted and recorded sEMG was observed across eight subjects, five muscles, and four weights. In addition, the time domain features like integrated EMG (IEMG), root-mean-square (RMS) value, and variance of EMG (VEMG) were compared between forecasted and recorded sEMG (fatigue state), which yielded an average MAPE of 8%, 19.2%, and 31.7%, across eight subjects, five muscles, and four weights, for (IEMG and MAV), RMS, and (VEMG and SSI) respectively. The results encourage combining the proposed approach with wearable technology for forecasting fatigue-induced sEMG to drive stimulation devices like FES and robotic devices. </p>

Quantitative assessment of muscle fatigue during rowing ergometer exercise using wavelet analysis of surface electromyography (sEMG)

“…In some cases, positive regression coefficients for MDF were observed, which may be due to various limitations. One such limitation is the individual characteristics of the participants, which inherently vary among them (for example: displacement and modification of the volume of the muscle being analysed influence on EMG measure (Romero and Gual, 2010) or sweating during the long duration of exercise (Bala and Joshi, 2022)). Furthermore, distortions and interference of the sEMG signal, which can result from, e.g., crosstalk of EMG signals of adjacent muscles, cannot be eliminated during the data analysis phase, which can also affect these results (Zhang et al, 2022).…”

“…In dynamic movement, changes in sensor positioning may occur and cause sEMG signal artifacts (Perpetuini et al, 2023), which adds an additional layer of limitation to this type of measurement. Moreover, investigations into the prediction of fatigue-induced electromyographic signals are also of interest (Bala and Joshi, 2022). One study explored this by analyzing initial sEMG recorded over a shorter time period during bicep flexion with a load.…”