Relationship between EMG features and force in orbicularis oris muscle

Abstract

BACKGROUND:

Lip incompetence resulting from mouth breathing is a common clinical manifestation, while there are no definite indicators of amplitude and intensity of muscle functional training in clinical practice, which leads to unsatisfactory training results.

OBJECTIVE:

The aim was to quantify the relationship between electromyography (EMG) and force in orbicularis oris muscle, so that the indicators of muscle functional training can be evaluated using EMG signals, so as to improve the training effects.

METHODS:

The EMG and the force signals of orbicularis oris muscle from 0% to 100% MVC within 5 s in twelve healthy subjects (six males and six females; age, 25 $\pm$ 2 years; mass, 60 $\pm$ 15 kg) were recorded simultaneously for three trials. Four EMG features consisting of RMS, WAMP, SampEn and FuzzyEn were analyzed. The regression analyses were performed using first-order and third-order polynomial model.

RESULTS:

There were high correlations between the four EMG features and muscle force with the two models. The third-order model yielded a higher coefficient of determination ( $R^{2}$ ) than the linear model ( $p<$ 0.001) and the result of FuzzyEn ( $R^{2}$ : 0.884 $\pm$ 0.059) was the highest in the four features.

CONCLUSION:

The third-order model with FuzzyEn of EMG signals may be used to guide the muscle functional training.

Keywords

Orbicularis oris muscle surface electromyography muscle force FuzzyEn

1. Introduction

Starting with Inman et al. [1] in 1952, a large number of studies on the relationship between electromyography (EMG) signals and muscle force have been conducted by domestic and foreign scholars, because it has a wide range of applications in the fields of sports science, rehabilitation medicine and biomechanics. The EMG signals and muscle contraction are regarded as both electrical and mechanical outcomes of muscular activities. Therefore, the relationship between EMG and muscle contraction (described as force or torque alternatively) offers a useful tool for examining muscle functions and deficits with wide range of biomedical and biomechanical applications [2]. Surface electromyography (sEMG) is a kind of complex bioelectrical signal, which contains rich muscle function and state information, and can reflect the activation degree of skeletal muscle and be conveniently collected non-invasively. Therefore, sEMG is considered as the most potential indirect muscle force measurement method [3].

Many previous studies have reported complex relationships between the EMG signals and muscle force. Some studies have shown a linear relation between the amplitude of sEMG and the force in biceps brachii muscle [4, 5]. However, some scholars have suggested that the EMG-force relation was more likely to be nonlinear in biceps brachii muscle and quadriceps femoris muscle [6, 7]. Furthermore, some scholars have also found that the EMG-force relationship was affected by multiple factors, such as sensor placement [5, 8], EMG features [2, 9] and force profiles [9, 10]. In general, the EMG signals are analyzed with time-domain amplitude-associated features and frequency-domain features including root mean square (RMS) [11, 12], Wilson amplitude (WAMP) [9, 13], peak value [14, 15], envelope [16, 17], mean frequency (MF) [18] and power spectral analysis [19] etc., while nonlinear complexity domain features have also been widely used in recent years such as sample entropy (SampEn) [20, 21] and fuzzy entropy (FuzzyEn) [2].

Mouth breathing is a common disease in children mostly caused by adenotonsillar hypertrophy. Chronic bad breathing patterns accompanied by lips muscle deficiency eventually could lead to malocclusion such as mandibular retrognathism, dental protrusion, high angle and so on [22]. In clinical practice there are some training methods such as pursing lips, boing sounds and hanging heavy loads to improve lips muscle force, and efficacy has been observed. But there are no definite indicators of amplitude, intensity and time of muscle functional training, which leads to poor patient compliance and inefficient training. Some have tried to quantify the training with definite pressure [23], while no pressure sensing is performed for most of the training movements. Some have tried to compare the myoelectric activity of the lips instead of pressure [24, 25], but to our knowledge, none of studies on quantifying the relationship between EMG and force in orbicularis oris muscle have been reported. Hence, studying the relationship between EMG and force in orbicularis oris muscle could provide an alternative and useful way to represent the lips muscle strength, so that the amplitude, intensity and time of muscle functional training can be evaluated using EMG signals, so as to improve the training results.

In order to quantify the relationship between EMG and force in orbicularis oris muscle twelve healthy subjects were recruited for this study. The EMG signals were recorded with an 8-channel linear electrode array from the upper orbicularis oris muscle of subjects, at the same time a thin film pressure sensor was put between the upper and lower lips to record the force signals as an approximate estimate of muscle force. Time-domain and nonlinear complexity domain features were both employed in this study to interpret the recorded EMG signals and to evaluate the EMG-force relationship in orbicularis oris muscle.

2. Methods

2.1 Participants

Twelve healthy subjects (six males and six females; age, 25 $\pm$ 2 years; mass, 60 $\pm$ 15 kg) without any known history of muscular or neural disorder participated in this experiment. None of the subjects had a history of musculoskeletal pain or injury in their orbicularis oris muscle. The experiment protocol was approved by the Ethics Committee of Shanghai University of China (Approval No. ECSHU 2021-180). Written, signed consents were obtained from all subjects prior to any procedure of the experiment.

2.2 Data collection

According to a previous study [26] on the optimal interelectrode distance and placement of surface EMG electrodes for orbicularis oris muscle, an 8-channel linear electrode array (OT Bioelettronica, Italy) with an interelectrode distance of 5 mm was used for EMG recording, and the electrode array was aligned to the muscle fiber direction. The subjects were seated in a chair with the experiment setup in front of them. The beards and hair around the lips were removed before the experiment and the target skin area was cleaned thoroughly with alcohol wipes to reduce the skin-electrode impendence. Then the electrode array was attached to the right muscle belly of the upper orbicularis oris muscle using double-sided adhesive tape with regular perforation patterns filled with conductive gel (Ten 20, Weaver and Company, USA), which was close to the facial midline in the horizontal dimension. A wristband was used as a common reference electrode. The EMG amplifier (EMG-USB2 $+$ , OT Bioelettronica, Italy) with a built-in bandpass filter of 10–500 Hz was used for recording the sEMG signals in a bipolar configuration at a sampling rate of 2048 Hz with a gain of 1000. Meanwhile, a thin film pressure sensor (FSR402, Shenzhen Weixin, China) was put between the upper and lower lips (not touch the teeth) and the auxiliary channel of the EMG amplifier was used to record force signals. The force signals were also sampled at 2048 Hz and amplified with a factor of 1000. The detailed information about the acquisition equipment and sensors is shown in Fig. 1.

Figure 1.

The side view and front view of the experimental setup. An 8-channel linear electrode array was put on the right muscle belly of the upper orbicularis oris muscle. A thin film pressure sensor was put between the upper and lower lips within a disposable waterproof plastic film.

Three trials of maximal voluntary contraction (MVC) of orbicularis oris muscle were first performed for each subject with a 3 minutes rest between the trials. The subjects were verbally encouraged to increase their muscle strength but not to use their teeth. Peak forces were determined for all trials from the data within a 0.25 s window centred on the overall maximum value. Within this window, the mean value was treated as the ‘peak’ force and recorded for each trial [27]. The maximum of the three replications was extracted for each subject and used to normalize the recorded forces signals for subsequent analyses.

Figure 2.

A bell-shaped force profile of 5 s.

Afterwards the subjects were asked to purse lips with a bell-shaped force profile of 5 s. The level of force spanned from 0% to 100% MVC then back to 0% MVC (Fig. 2). All subjects were provided with adequate time to practice matching the profile before the actual recording. The subjects were asked to perform three trials, so as to produce a sufficient amount of data. A 3 minutes rest was allowed between two consecutive trials for each subject in order to eliminate the potential effect of mental or muscular fatigue. The data from all channels could be displayed and recorded on the laptop screen in real time during the experiment.

The EMG and force signals were recorded simultaneously through EMG-USB2 $+$ amplifier and were stored in OTB format for offline analysis. Then the data was converted to MAT format and analyzed with Matlab (The Mathworks, Inc., Natick, MA, USA).

2.3 Data preprocessing and features calculation

2.3.1 Data segmentation and channel selection

Data preprocessing was necessary in order to reduce noise and obtain pure signals. The EMG signals were first notched by 50 Hz and then filtered by a zero-lag fourth-order Butterworth band pass filter (10 Hz to 500 Hz) to remove possible low frequency motion artifacts and high frequency interference. The force signals were filtered by a zero-lag second-order Butterworth low pass filter (10 Hz). In each trial, according to force signals with the ascent segment from 0% to 100% MVC, the EMG signals over the same period of time were selected for further analysis. Figure 3 showed an example for the data segment selection from subject 7. A sliding window was applied to the EMG signals with a width of 150 ms and a step of 37 ms, which was based on a previous study [28]. The EMG features were calculated for each of these windows. At the same time the same sliding window was applied to the force signals to calculate the average force magnitude.

Figure 3.

An example for the segment selection of EMG and force signals from subject 7. The data between the two red lines was selected for further analysis.

EMG signals were recorded in a bipolar configuration, which meant that the 8-channel linear electrode array produced 7 bipolar channels by subtracting each pair of adjacent channels along the muscle fibers, and the pattern of amplitude variation versus force might be different from channel to channel with force increasing [29]. Therefore, for a more robust assessment of the EMG-force relationship, all EMG features described below in this paper were extracted from the channel with the highest EMG amplitude, which can be regarded as the position on muscle belly farther away from the innervation zone, thereby better reflecting the muscle activity [20]. Based on the assumption that the channel with the highest EMG amplitude was consistent across multiple trials for the same muscle, a majority voting strategy was subsequently performed to determine a representative channel that globally yielded the highest EMG amplitude among all the channels, over multiple data segments for each subject [20].

2.3.2 Parameters of EMG features

The time-domain characteristics of the EMG signals were chosen as RMS and WAMP, which had been widely used in previous studies. WAMP is used to count the number of times that the signal amplitude exceeds a predefined threshold, is an indicator of firing of motor unit action potential (MUAP) and therefore is an indicator of muscle contraction level [30]. The threshold for WAMP was set to 0.06 mV in this paper.

The nonlinear complexity domain features chosen for the EMG signal were SampEn and FuzzyEn, which are calculated similarly and involved three parameters, namely the embedding dimension $m$ , the signal length $N$ and the tolerance $r$ . Generally, the embedding dimension $m$ is set to 2 empirically, and the signal length $N$ is determined by $10^{m}\sim 10^{m}$ [31]. Although this length is feasible for entropy calculation, some literature recommended to process sufficiently long signal to avoid notable bias [20]. In this paper, the force rise segment of 0% to 100% MVC corresponding to the EMG data segment was chosen to calculate the both entropies with at least 2 s of the data, which can be considered sufficient. Usually, the tolerance $r$ is defined as:

$\displaystyle r=k*\textit{std}(x)$ (1)

where $\textit{std}(x)$ represents the standard deviation of the signals. Empirically, 0.1–0.3 are all reasonable for the parameter $k$ . The choice of the tolerance $r$ should be considered specially. A small tolerance would make the entropy interfered remarkably by noise, and inversely, a large one might cause detailed information loss [29]. Zhang Xu et al. [20] calculated entropy with both global and local tolerance schemes to compare the EMG complexity-torque relation between two sides of stroke subjects. Despite of the relative variation in entropy values, they considered the choice of either tolerance scheme did not affect the main findings of their study. Given that the global tolerance takes into account the signal amplitude when measuring the signal complexity, in this paper the entropies of EMG signals were calculated with global tolerance and $k$ was set to 0.3.

2.4 Evaluation of EMG-force relationship

2.4.1 Modeling of EMG-force relationship

The regression analysis was performed in order to explain the relationship between each of the four EMG analytic features and the muscle force respectively. Since previous studies reported linear [4, 5] and nonlinear [6, 7] relation (e.g. third-order polynomial [13, 32]) between EMG and force, both the first-order linear model and third-order polynomial models were employed in this study, as described in the Eqs (2) and (3):

$\displaystyle f(x)=a*x+b$ (2) $\displaystyle f(x)=c*x^{3}+d*x^{2}+e*x+f$ (3)

Here, $a$ , $b$ , $c$ , $d$ , $e$ and $f$ were six coefficients in two regression models.

2.4.2 Evaluation of EMG-force relationship

Coefficient of determination ( $R^{2}$ ) for the regression was utilized to evaluate the fitting across all four EMG features. In order to examine the effect of the feature and the regression model on the EMG-force relationship, a two-way repeated-measure ANOVA (analysis of variance) was applied on the $R^{2}$ value derived from the regression analysis, with the model (two levels: linear and third-order) and feature (four levels: RMS, WAMP, SampEn and FuzzyEn) both considered as within-subjects factors. The statistical analysis was conducted by using the SPSS software (ver. 21.0, SPSS Inc., Chicago, IL, USA). Significance level was set to $p<$ 0.05. The Bonferroni-Dunn adjustment was used for multiple comparisons.

The k-fold cross validation [33] was introduced to evaluate the performance of the model, and k was set to three in this paper. For all subjects, two complete trials were used for training and the third trial was used for testing. This process was iterated such that each trial had been used for testing. Root mean squared error (RMSE) between the recorded force and the estimated force across the three cross-validated test trials were calculated.

3. Results

Representative examples of characterizing EMG-force relationship of subject 7 using both linear and third-order regression were shown in Fig. 4, where the EMG was described by RMS, WAMP, SampEn and FuzzyEn in a, b, c, and d, respectively. The regression analyses confirmed high correlation with $R^{2}$ above 0.8 between the EMG features and the muscle force with two models in Fig. 4. It could be observed that the third-order model yielded a higher $R^{2}$ than the linear model for all four features, with similar results obtained in the fitting of other subjects in this study. Figure 5 was a box plot of the $R^{2}$ derived from all four features using linear and third-order regression methods. It can be clearly found that the median of $R^{2}$ with the FuzzyEn was higher than the other three features in both regression models.

Figure 4.

Representative examples of characterizing EMG-force relationship of subject 7 using linear regression (Red line) and third-order regression (Blue line). The four graphs represented four EMG features: (a) RMS; (b) WAMP; (c) SampEn; (d) FuzzyEn. Individual data points were shown by black stars.

Figure 5.

Boxplot graphs of the $R^{2}$ derived from all four features of all subjects using linear and third-order regression methods, RMS: blue; WAMP: carmine; SampEn: black; FuzzyEn: green.

The two-way ANOVA (with models and features) showed that the third-order model achieved a significantly higher $R^{2}$ than the simple linear model ( $p<$ 0.001). In addition, the $R^{2}$ derived from FuzzyEn (0.884 $\pm$ 0.059) with third-order model was higher than that from RMS (0.854 $\pm$ 0.069) and WAMP (0.857 $\pm$ 0.079) ( $p<$ 0.05), and was also higher than the $R^{2}$ from SampEn (0.872 $\pm$ 0.062) ( $p>$ 0.05).

The RMSE of the third-order model was lower than the linear model from three-fold cross validation across all subjects ( $p<$ 0.05). The RMSE of FuzzyEn with the third-order model was 0.122 $\pm$ 0.048.

4. Discussion

The study evaluates the EMG-force relationship for the orbicularis oris muscle. The orbicularis oris muscle is the circular muscle around the rima oris, which is partly composed of fibers of other perioral muscle and it is structurally complex. The EMG recording from the skin may not uniquely record from the targeted muscle. Though the observed EMG signals might be influenced by the adjacent muscles via crosstalk, the degree of activation of orbicularis oris muscle still could be observed from the recorded EMG signals, because the regression analyses confirm high correlations between the EMG signals and the orbicularis oris muscle force with the two models across all subjects.

Fewer studies have been conducted on the EMG-force relationship in orbicularis oris muscle, and linear and nonlinear relationships have been reported for other muscles, such as the biceps brachii and the first dorsal interosseous muscle, while some researchers consider that the nonlinear model between the EMG and muscle force is more accurate. The linear and third-order regression analyses are employed in this study and both of the two models yield a high $R^{2}$ ( $R^{2}>$ 0.8). In addition, the third-order regression model performs better than the linear model ( $p<$ 0.001), which is consistent with the previous study [32]. The RMSE value of the three-fold cross validation with the third-order model is acceptable and consistent with the previous results [8], which also shows that is better than the linear model ( $p<$ 0.05).

There are four features to interpret the EMG signals in this study. Using EMG amplitude is a straightforward and practical solution for estimating the muscle force [34]. Among a category of methods relying on macroscopic EMG amplitude-associated features, both the RMS and WAMP features are implemented in this study and achieve acceptable performance. This confirms the previous finding that the EMG signal amplitude generally had the same trend as the muscle force [35]. Considering the nonlinear and non-stationary properties of EMG signals, interpreting data with conventional time-domain features may unavoidably have limitations. Two different complexity measures are also implemented in this study, namely SampEn and FuzzyEn. Both of them are highly correlated with the orbicularis oris muscle force and increase monotonically with the muscle force. Such a finding can be attributed into neurophysiological processes of muscle force production. By recruiting more motor units and increasing their firing rates, the appearance of more overlapped motor unit action potentials in the EMG signal leads to its increased level of complexity evaluated by entropy measures [2]. The performance of FuzzyEn is better than SampEn, but there is no significant difference between them.

Finally, the primary limitation of the study is the limited number of recruited healthy subjects, which leads to insufficient statistical power in some comparisons. However, this does not affect the preliminary conclusions obtained in this paper. Our future work will focus on extensive experiments with more healthy subjects and subjects with abnormal lip function, as well as on the application and comparison of some EMG complex analysis methods.

5. Conclusions

In this paper, the EMG-force relationship in orbicularis oris muscle is studied using linear and third-order regression models, where the EMG signals are interpreted by amplitude-associated features and entropy features. The regression analyses using two models and four features all yield a relatively high $R^{2}$ ( $R^{2}>$ 0.8). The $R^{2}$ of the FuzzyEn is the highest in the four features. The statistical analysis shows that the third-order model achieves a significantly higher $R^{2}$ and a lower RMSE value than the simple linear model ( $p<$ 0.05).

Our study confirms high correlations between the EMG features and the orbicularis oris muscle force, and the third-order model with FuzzyEn represents the force better, which may be used to guide the muscle functional training for lip incompetence in clinical practice.

Footnotes

Acknowledgments

The authors would like to thank the Science and Technology Commission of Shanghai Municipality (grant number: 16441909000) and the China Oral Health Foundation (grant number: A2021-145) for the financial support and all volunteers who participated in the study and contributed their time.

Conflict of interest

None to report.

References

Inman

Ralston

Saunders

Feinstein

Wright

Jr. Relation of human electromyogram to muscular tension. Electroencephalogr Clin Neurophysiol. 1952 May; 4(2): 187-94. doi: 10.1016/0013-4694(52)90008-4.

Zhu

Zhang

Tang

Gao

Chen

. Re-evaluating electromyogram-force relation in healthy biceps brachii muscles using complexity measures. Entropy. 2017 Nov; 19(11): 624. doi: 10.3390/e19110624.

Huang

. Research and Implementation of Muscle Fatigue Analysis and Muscle Force Prediction based on Surface Electromyography. University of Science and Technology of China. 2014.

Fukuda

Echeimberg

Pompeu

Lucareli

PRG

Garbelotti

Gimenes

Apolinario

. Root Mean Square value of the electromyographic signal in the isometric torque of the quadriceps, hamstrings and brachial biceps muscles in female subjects. J Appl Res. 2010 Jan; 10(1): 32-39.

Ahamed

Sundaraj

Alqahtani

Altwijri

Ali

Islam

. EMG-force relationship during static contraction: Effects on sensor placement locations on biceps brachii muscle. Technol Health Care. 2014; 22(4): 505-13. doi: 10.3233/THC-140842.

Sbriccoli

Bazzucchi

Rosponi

Bernardi

De Vito

Felici

. Amplitude and spectral characteristics of biceps Brachii sEMG depend upon speed of isometric force generation. J Electromyogr Kinesiol. 2003 Apr; 13(2): 139-47. doi: 10.1016/s1050-6411(02)00098-6.

Bilodeau

Schindler-Ivens

Williams

Chandran

Sharma

. EMG frequency content changes with increasing force and during fatigue in the quadriceps femoris muscle of men and women. J Electromyogr Kinesiol. 2003 Feb; 13(1): 83-92. doi: 10.1016/s1050-6411(02)00050-0.

DiDomenico

Nussbaum

. Estimation of forces exerted by the fingers using standardised surface electromyography from the forearm. Ergonomics. 2008 Jun; 51(6): 858-71. doi: 10.1080/00140130801915980.

Bøg

Erkocevic

Niemeier

Mathiesen

. Investigation of the Linear Relationship between Grasping Force and Features of Intramuscular EMG. 15th NBC on Biomedical Engineering and Medical Physics (NBC 2011), Aalborg, Denmark. 2011. pp. 121-124. doi: 10.1007/978-3-642-21683-1_30.

10.

Hoozemans

van Dieën

. Prediction of handgrip forces using surface EMG of forearm muscles. J Electromyogr Kinesiol. 2005 Aug; 15(4): 358-66. doi: 10.1016/j.jelekin.2004.09.001.

11.

Turingan

Cruz

FRG

Esporlas

. Wireless Measuring System with Wavelet Denoising for Filipino Maximum Bite Force and Orofacial Muscle Surface EMG. 2018 IEEE 10th International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment and Management (HNICEM), Baguio City, Philippines. 2018. pp. 1-5. doi: 10.1109/HNICEM.2018.8666371.

12.

Litchy

Evertz

Kaufman

. Evaluating skeletal muscle electromechanical delay with intramuscular pressure. J Biomech. 2018 Jul 25; 76: 181-188. doi: 10.1016/j.jbiomech.2018.05.029.

13.

Kamavuako

Rosenvang

. Hysteresis in the electromyography-force relationship: Toward an optimal model for the estimation of force. Muscle Nerve. 2012 Nov; 46(5): 755-8. doi: 10.1002/mus.23393.

14.

Choi

Lee

Kim

. A Study on Estimation of Joint Force Through Isometric Index Finger Abduction with the Help of SEMG Peaks for Biomedical Applications. IEEE Trans Cybern. 2016 Jan; 46(1): 2-8. doi: 10.1109/TCYB.2014.2386856.

15.

Kim

. Force estimation in fatigue condition using a muscle-twitch model during isometric finger contraction. Med Eng Phys. 2017 Dec; 50: 103-108. doi: 10.1016/j.medengphy.2017.10.002.

16.

Huang

Chen

Cao

Qiu

Zhang

. An isometric muscle force estimation framework based on a high-density surface EMG array and an NMF algorithm. J Neural Eng. 2017 Aug; 14(4): 046005. doi: 10.1088/1741-2552/aa63ba.

17.

Chen

Yuan

Cao

Zhang

Chen

. A SEMG-force estimation framework based on a fast orthogonal search method coupled with factorization algorithms. Sensors (Basel). 2018 Jul 11; 18(7): 2238. doi: 10.3390/s18072238.

18.

Karlsson

Gerdle

. Mean frequency and signal amplitude of the surface EMG of the quadriceps muscles increase with increasing torque – a study using the continuous wavelet transform. J Electromyogr Kinesiol. 2001 Apr; 11(2): 131-40. doi: 10.1016/s1050-6411(00)00046-8.

19.

Shin

Zhou

Niu

Liu

Rymer

. Power spectral analysis of surface electromyography (EMG) at matched contraction levels of the first dorsal interosseous muscle in stroke survivors. Clin Neurophysiol. 2014 May; 125(5): 988-94. doi: 10.1016/j.clinph.2013.09.044.

20.

Zhang

Wang

Chen

Zhou

. EMG-torque relation in chronic stroke: A novel EMG complexity representation with a linear electrode array. IEEE J Biomed Health Inform. 2017 Nov; 21(6): 1562-1572. doi: 10.1109/JBHI.2016.2626399.

21.

Kamavuako

Farina

Jensen

. Use of sample entropy extracted from intramuscular EMG signals for the estimation of force. 15th NBC on Biomedical Engineering and Medical Physics (NBC 2011), Aalborg, Denmark. 2011. pp. 125-128. doi: 10.1007/978-3-642-21683-1_31.

22.

Chen

Zhou

Jiang

. The training method for the orthodontic treatment of the lips muscle relaxation. Medical Science Journal of Central South China. 2011; 39(02): 198-200. doi: 10.15972/j.cnki.43-1509/r.2011.02.033.

23.

Park

Kwon

Choi

Kim

. Effect of orbicularis oris muscle training on muscle strength and lip closure function in patients with stroke and swallowing disorder. J Phys Ther Sci. 2018 Nov; 30(11): 1355-1356. doi: 10.1589/jpts.30.1355.

24.

Uysal

Yagci

Kara

Okkesim

. Influence of pre-orthodontic trainer treatment on the perioral and masticatory muscles in patients with Class II division 1 malocclusion. Eur J Orthod. 2012 Feb; 34(1): 96-101. doi: 10.1093/ejo/cjq169.

25.

Yoshizawa

Ohtsuka

Kaneko

Iida

. Assessment of hypoxic lip training for lip incompetence by electromyographic analysis of the orbicularis oris muscle. Am J Orthod Dentofacial Orthop. 2018 Dec; 154(6): 797-802. doi: 10.1016/j.ajodo.2018.02.009.

26.

Xiang

. Research and development of surface EMG signal detector of children’s orbicularis oris muscle. Shanghai University. 2021.

27.

Didomenico

Nussbaum

. Measurement and prediction of single and multi-digit finger strength. Ergonomics. 2003 Dec 15; 46(15): 1531-48. doi: 10.1080/0014013032000121660.

28.

. Onset detection of surface diaphragmatic electromyography and its application in human-computer interaction. Shanghai University. 2019.

29.

Jahanmiri-Nezhad

Suresh

Rymer

Zhou

. EMG-force relation in the first dorsal interosseous muscle of patients with amyotrophic lateral sclerosis. NeuroRehabilitation. 2014 Jan 1; 35(2): 307-14. doi: 10.3233/NRE-141125.

30.

Huang

Chen

. Development of a myoelectric discrimination system for a multi-degree prosthetic hand. Proceedings 1999 IEEE International Conference on Robotics and Automation. 1999. pp. 2392-2397. doi: 10.1109/ROBOT.1999.770463.

31.

Chen

Zhuang

Wang

. Measuring complexity using FuzzyEn, ApEn, and SampEn. Med Eng Phys. 2009 Jan; 31(1): 61-8. doi: 10.1016/j.medengphy.2008.04.005.

32.

Al Harrach

Carriou

Boudaoud

Laforet

Marin

. Analysis of the sEMG/force relationship using HD-sEMG technique and data fusion: A simulation study. Comput Biol Med. 2017 Apr 1; 83: 34-47. doi: 10.1016/j.compbiomed..

33.

Kamavuako

Scheme

Englehart

. Wrist torque estimation during simultaneous and continuously changing movements: Surface vs. untargeted intramuscular EMG. J Neurophysiol. 2013 Jun; 109(11): 2658-65. doi: 10.1152/jn.00086.2013.

34.

Zhang

Zhu

Chen

Zhou

. Muscle force estimation based on neural drive information from individual motor units. IEEE Trans Neural Syst Rehabil Eng. 2020 Dec; 28(12): 3148-3157. doi: 10.1109/TNSRE.2020.3042788.

35.

Staudenmann

Kingma

Daffertshofer

Stegeman

van Dieën

. Heterogeneity of muscle activation in relation to force direction: A multi-channel surface electromyography study on the triceps surae muscle. J Electromyogr Kinesiol. 2009 Oct; 19(5): 882-95. doi: 10.1016/j.jelekin.2008.04.013.