Table 2.

Summary of studies on electrically evoked contraction and evoked EMG-fatigue relationship.

Reference	Clinical Population and Study Design	Parameter Analysed	Outcome/Findings
Mizrahi et al., 1994 [3]	4 complete SCI individuals Surface stimulation was administered at the motor point to evoke isometric contraction of the quadriceps muscle for maximal stimulus with the SF of 20 Hz and PW of 0.25 ms Hip and knee angle were held constant at 90° and 30°, respectively, for each participant	PTP amplitude of M-wave	The correlation coefficient of PTP amplitude of eEMG and force was up to (r = 0.90) during fatigue. Circuit designed to suppress stimulation artefact led to another form of noise called electrode offset potential of higher magnitude than M-wave.
Erfanian et al., 1996 [54]	1 complete SCI individual Six minutes of a sustained percutaneous stimulation was used to activate vastus lateralis under isometric condition at 30° flexion and 0° extension knee angles The SF was 20 Hz and constant amplitude of 20 mA.	PA, MDF	PA and the power spectrum increased during potentiation, decreased during fatigue and increased again during maximal fatigue Postactivation potentiation, fatigue, and maximal fatigue states were manifested after a prolonged stimulation.
Tetapac et al., 1997 [40]	4 complete SCI and 2 healthy individuals The subject were considered untrained for FES contraction Surface stimulation with the SF of 25 Hz, and PW of 0.25 ms was administered at the motor point of the wrist flexor for maximal stimulus under isometric muscle action. Information about FES-induced fatigue was derived from the force decline as shown on torque versus EMG curve.	MDF, MAV, PTP, RMS	The drop in MDF gave an indication of fatigue due to the neuromuscular propagation. Force level varied between 62% and 96% between initial (non-fatigue) and end of recovery stage. Implications on the dynamic contraction and prolonged fatigue were not described.
Chesler & Durfee, 1997 [55]	3 SCI and 20 healthy individuals 1 h per day, 5 days a week of an isometric quadriceps muscle strengthening by a transcutaneous stimulation with the SF of 30 Hz and PW of 0.3 ms was administered	RMS, MF, MAV	Noiseless eEMG was difficult to obtain, thus, limited the usage in FES practical application. Amplitude based parameters of eEMG were more relevant than the frequency based as indicator of muscle fatigue.
Chen & Yu, 1997 [49]	4 Complete SCI individuals Surface electrical stimulation of SF of 20 Hz and PW of 0.3 ms and surface eEMG acquisition were adopted Fatigue protocol was conducted on cycle ergometer. Only quadriceps muscles were stimulated in isolated stage while quadriceps and hamstring were stimulated in reciprocal fashion.	PTP amplitude	During continuous and intermittent stimulation there were positive correlation between PTP of eEMG and muscle force (r = 0.94) and (r = 0.78), respectively. The decrease of PTP of EMG from a maximum value of 0.66 mV to an asymptotic value of 0.5 mV signified the metric of fatigue. The investigators suggested that evoked EMG may not be sensitive to fatigue during dynamic contraction because of the larger inflection time and the time constant of the PTP of evoked EMG.
Yu et al., 1999 [56]	5 SCI individual with lesion between C7-T11 Transcutaneous stimulation of the quadriceps was used with monophasic WF of frequency- 20 Hz, PW- 300 ms and maximum current was 120 mA eEMG was obtained from the muscle belly of the quadriceps To induce fatigue; an isometric and dynamic muscle actions were administered between 30° and 110° of knee flexion at 30°/s for dynamic contraction.	PTP amplitude, RTP, PTP duration, and torque	During fatigue; the decline in the PTP was positively correlated with the decline in the force output (r = 0.88, p < 0.05) while the temporal features RTP and PTP duration were negatively correlated with the decline in the torque (r = −0.74 and −0.73, p < 0.05) respectively. The decaying rate of the temporal feature and the torque output were slower in isometric contraction that in dynamic, i.e., dynamic contraction is more susceptible to muscle fatigue.
Heasman et al., 2000 [1]	2 SCI individuals Implanted stimulator was used to activate EDC, FPL, EPL muscles with peripheral nerve stimulation of PW modulation (0–0.2 ms), constant current of 20 mA and SF of 12 Hz. Load cell was used to measure the isometric muscle force Protocol was conducted once per day, and repeated once every four weeks on a separate occasion for each participant	(PTP, RMS, SPA) amplitude, MNF	SPA and RMS of M-wave demonstrated the highest correlation (r = 0.88) to force during non-fatigue or fatigue state. M-wave parameters indicated muscle electrical activation, but were relatively invariant to muscle fatigue. Non-isometric contraction was not investigated.
Estigoni et al., 2011 [51]	8 SCI individuals FES-cycling of 15 min duration with 5 min recovery time performed by each participant 2–3 times per week for at least 6 weeks of FES-cycling sessions before the test Transcutaneous SF of 25 Hz and PW of 0.3 ms was administered to activate quadriceps muscle eEMG was carefully acquired from rectus femoris muscle	PTP amplitude	Variation in magnitude of M-wave changes compared to torque changes disallowed statistical modelling understanding of the fatigue effect generated by M-wave curve. eEMG could not predict the decrement of muscle torque during fatiguing FES-cycling.
Li et al., EMG [30]	5 SCI individuals (3-T6, 1-C5, 1-C7) Surface stimulation of the right triceps surae was delivered to plantarflex the ankle joint with the constant frequency (30 Hz) and constant PW (450 μs). Isometric angle planter-flexion torque was acquired. Each subject performed both fatigue-inducing (1 s ramp up, 2 s plateau, 1 s ramp down then 2 s rest) and random tests. Maximum stimulation amplitude was set at the point where the torque became saturated for each subject.	MAV, Torque	The NARX-RNN demonstrated a robust identification performance while keeping its accuracy and stability. Future work to verify the performance of the model on the adaptive closed loop FES control for dynamic motion based on eEMG and angle-velocity sensing. Due to the variability in the subjects' level of neurological lesion, subject-specific model may be more suitable.

Abbreviations: PW: Pulse width; SF: Stimulation frequency; PTP: Peak to peak amplitude; PA: Peak amplitude; RMS: Root mean square; MDF: Median frequency; MAV: Mean absolute value; SPA: Second phase area (area under the curve of second phase of average M- wave); MNF: Mean frequency; WF: wave form; PW: Pulse duration; EDC: Extensor Digitorum Communis; EPL: Extensor Pollicis Longus; FPL- Flexor Pollicis Longus; NARX-Nonlinear autoregressive exogenous model; RNN-Recursive neural network.