Wavelet analysis of the emg signal in application to assessment of load and fatigue of the musculoskeletal system

Bartuzi, Pawel

Department of Ergonomics / Central Institute for Labour Protection - National Research Institute / Czerniakowska 16 / 00-701 Warsaw, Poland+48 22 623 37 81 / pabar@ciop.plRoman-Liu, DanutaDepartment of Ergonomics / Central Institute for Labour Protection - National Research Institute / Czerniakowska 16 / 00-701 Warsaw, Poland+48 22 623 32 75 / daliu@ciop.pl

ABSTRACT

The aim of the study was to compare the efficacy of the EMG signal parameters calculated on the basis of frequency and time-frequency analysis in the evaluation of muscle fatigue, taking into account the level of load of the muscle. The EMG signal was measured from triceps brachii (TB), in static conditions, on three levels of load, till the occurrence of fatigue. EMG signal analysis in frequency (Fourier transform) and time- frequency (continuous and discrete wavelet transform) domain was carried out. The parameters determined on the basis of wavelet transform show a greater dependence on fatigue, than parameters calculated on the basis of Fourier transform. Additionally wavelet parameters, in contrast to Fourier parameters, do not require the adoption of simplifying assumptions about the stationarity of the tested signal. The results will allow to more accurate assessment of muscle fatigue during work.

Key Words.

Muscle fatigue, muscle load, EMG, wavelet

INTRODUCTION

Disorders of the musculoskeletal system can be caused by excessive muscle load and fatigue (Visser and van Dieën 2006; Roman-Liu and Konarska, 2009). This means that the reduction of the load and fatigue in the working conditions may be an important step in reducing the occurence of musculoskeletal disorders. One of the commonly used methods to access the load and fatigue from the muscles is a surface electromyography (EMG) (Piscione and Gamet 2006; Bartuzi et al. 2010).

In order to assess muscle load during various activities the analysis of amplitude (RMS

- root mean square) of the EMG signal is carried out (Troiano et al. 2008; Ioi et al. 2008; De Luca 1997). Muscle fatigue is visible in the values of the EMG signalparameters such as root mean square (RMS), mean power frequency (MPF) or median frequency (MF) of the power spectrum. Muscle fatigue causes an increase in RMS and shifts the power spectral density (PSD) of the EMG signal towards relatively lower frequencies, which is manifested by a decrease in the values of MF and MPF (Gates and

Dingwell, 2008; Yamada et al. 2008; Barandun et al. 2009; Roman-Liu et al. 2004). The parameters of the EMG signal depend not only on muscle load and fatigue, but also on recruitment of new, non-fatigued motor units (Boonstra et al. 2008; Dimitrovaet al. 2009; Klaver-Król et al. 2010), proportion of specific fibres in the muscle (Bilodeau et al. 2003; Gerdle et al. 2000; Larsson et al. 2006; Pincivero et al. 2000) and fatty tissue content (Bartuzi et al. 2010; Farina and Mesin, 2005; Petrofsky 2008). There are various methods of the EMG signal analysis. For many years the commonlyused methods based on time and frequency analysis. In recent years, time-frequency methods are used in the EMG signal analysis. One of the time-frequency methods of the EMG signal analysis is the wavelet transform (Daubechies, 1992). The results of research indicate that the time-frequency methods can be used in the analysis ofmuscle load and fatigue (So et al, 2009; Flanders, 2002; Pope et al. 2000; von Tscharner, 2000). Authors of many studies indicate that the time-frequency methods provide more accurate results concerning the assessment of muscle load and fatiguethan frequency methods (Larivière et al. 2008; Barandun et al. 2009; von Tscharner 2000). In addition, for the time-frequency methods is not necessary to adopt simplifying assumptions about the stationarity of a signal.

The aim of the study was to compare the efficacy of the EMG signal parameterscalculated on the basis of frequency and time-frequency analysis in the evaluation of muscle fatigue, taking into account the level of load of the muscle.

METHODS

1 Participants

There were 15 male participants. The participants were physically active, and they did not suffer from any musculoskeletal diseases. Mean (standard deviation) age, body weight and height were, respectively, 21.9 years (1.4); 73.6 kg (3.9); 180.1 cm (3.2). The protocol of the study was approved by the local ethics committee. Each participant read and signed an informed consent form prior to the study.

2EMG Signal measurement

The experimental studies with usage of surface electromyography were carried out. Raw EMG signal from right triceps brachii caput laterale (TB) in isometric conditions was measured and registered with a Bagnoli-16 device (Delsys, USA). Data were sampled with a 16-bit DAQCard 6036E A/D card (National Instruments, USA). The signal’s sampling frequency was 4 kHz. The bandwidth of Bagnoli-16 was 20–450 Hz (±10%), bandwidth roll-off 80 dB/decade, overall noise ≤1.2 µV (RMS, RTI) and EMGamplification 1000. The EMG signal was recorded with EMG Works 3.5 software and double differential surface electrodes DE-3.1 (Delsys, USA). Those electrodes were used to reduce the risk of crosstalk (De Luca, 1997; Farina et al. 2002). Contact material of the sensor was 99.9% Ag, contact dimension 10 × 1 mm and contact spacing 10 mm, so the detection area was 200 mm2 (including both pairs of contacts). Input impedance of the sensor was over 1015Ω //0.2 pF, noise 1.2 µV (RMS, RTI), CMRR -92 dB and preamplifier gain 10.

First the skin was shaved and disinfected with a cotton swab soaked in alcohol. The electrodes were also disinfected with alcohol. They were fixed to the skin with a dedicated double-sided adhesive tape; thus, there was good contact between the electrodes and the skin throughout the study. To avoid signal artefacts no gel was used (Roy et al. 2007). After affixing the electrodes, participants were asked to relax and to activate the muscles in question to provide a high signal-to-noise ratio (De Luca 2002; Hermens et al. 2000). To do that they straightened the elbow joint against resistance. Differences in the amplitude of the EMG signal between the state of activation and relaxation were visually checked. If the difference was not clear, the contact between the skin and the electrode was improved.

The electrodes were located on the skin according to the SENIAM guidelines (Hermens et al. 1999; 2000) and Perotto (2005). The electrodes were placed at one-half on the line between the posterior crista of the acromion and the olecranon at 2 finger widths lateral to the line (the electrodes were oriented in the direction of the line between the posterior crista of the acromion and the olecranon).

3.Force measurement

The levels of load (50%MVC, 60%MVC and 70%MVC) were determined on the basis of the value of external force measured with the usage of the dynamometer. The dynamometer in conjunction with a transducer allows converting an applied force to an electrical signal and the presentation of changes in the force values during a test. The software CPSv_2.0 was used to visualise and measure the force. The visualisation of force enabled the subject observation and maintaining the exerted force on a constant level.

4.Protocol

In order to activate tested muscle the measurement stand USMS was used. Figure 1 shows the activation pattern of the muscle TB on the USMS stand.

Fig 1. The activation of TB muscle on the USMS stand

During measurements subjects were standing upright, with their right upper limb flexed in the elbow at the angle of 90° (fig. 1). The flexion angle of the shoulder was also 90°. The task of the subjects was to straighten the elbow joint against resistance. The measurements were performed in two stages. The first stage of measurements included the registration of the EMG signal and the external force at the maximum effort (maximal voluntary contraction, MVC). In one measurement the subjects developed two efforts, the larger was chosen as a MVC. The measurement at the maximum efforts took 10 seconds.

The second stage of measurements was performed for the same body posture as during tests in the first stage. The EMG signal was recorded, while subject maintained the constant force value during the tests at the levels of load 50%, 60% and 70%MVC. The order of tests for individual subjects were different and determined on the basis of the tables of random numbers.

Measurement of the EMG signal at each level of load lasted up to the point where the subject felt the exhaustion, or to the noticeable decrease in external force, which wasdisplayed on the monitor as a graph, of more than 20% of set level, for at least 2 consecutive seconds. Between tests there was a 15 minutes break for recovery.

SIGNAL ANALYSIS

1. Analysed fragments

The parameters which characterize the EMG signal registered from TB were computed with Matlab (version R2009) software. Two-second fragments of the EMG signal with the most stable values of the amplitude were analysed.

For each tested person fragments corresponding to the beginning (B) and the end (E) parts of trial on each of the levels of load (50%, 60% and 70%MVC) were chosen.

Fragments B correspond to the first two seconds of the load (after stabilization of the EMG signal amplitude), while E fragments correspond to the last two seconds of the load, in which the values of the EMG amplitude were stable. The differences in valuesof the EMG signal parameters between fragments B and E allow the evaluation of the impact of fatigue on the EMG signal characteristics. On the basis of selected fragments the parameters of the EMG signal were determined.

The duration of muscle activation (and EMG signal registration) at a given level of load(50%, 60% and 70%MVC) was different and depended on the load level and the endurance of the subject. Comparison of fatigue between different levels of load is possible only if the duration of the load in each of the analysed cases is the same. Therefore, the EMG signal parameters determined from the E fragments were normalized in time to get the values independent of the duration of the measurement. For the assessment of fatigue constant time of load (Tm) for all measurements (all load levels and all subjects) was adopted. In order to normalize values of the parameters for E fragments, on the basis of the trend line, values of the parameters for time Tm were calculated.

2.Paremeters of the frequency analysis

Fast Fourier transform (1 s, 4000 samples, Hanning window; 50% overlap) was used for spectral analysis of the selected fragments of the EMG signal registered in each test. As each window contained 4000 samples and the sampling frequency was 4 kHz, a spectral resolution of 1 Hz was obtained. Power spectral density (PSD) was estimated from the Fourier transforms with the Welch method and expressed with periodograms (Welch, 1967). On the basis of the periodograms MPF parameter was calculated

(equation 1).

                                                                                (1)

where S(f) is the value of density function for the frequency f, calculated by FFT.

3.Parameters of the time-frequency analysis

In order to determine the parameters in time-frequency domain, continuous wavelet transform (CWT) and discrete wavelet transform (DWT) were carried out. Wavelet function from Daubechies family (db5) was used.

In the case of CWT, wavelet coefficients for the 16 scales, representing the EMG signal in the frequency range from 19 Hz to 675 Hz, were calculated. Scales with the following numbers were selected: 6, 7, 9, 11, 13, 16, 20, 25, 31, 38, 47, 58, 69, 81,95, 109.

On the basis of CWT, for each of the selected scales, wavelet coefficients were obtained. Based on the wavelet coefficients from all scales the scalograms were obtained. From the scalograms the mean frequency CMPF was determined (Hostens and Ramon, 2005; Hostens et al. 2004):

                                                                          (2)

In equation (2) SC means scalogram, which is equivalent to the periodogram obtained on the basis of the Fourier transform, s is the scale (ls - the lowest scale, hs - the highest scale).

EMG signal analysis using discrete wavelet transform (DWT) was carried out at threelevels of decomposition. At each level of decomposition the approximation of the signal was analysed. Individual approximations are a representation of the EMG signal at a particular frequency band. Raw EMG signal contains frequencies in the range 0-2000 Hz (fs=4kHz). On the first level of decomposition approximation (A1) represents the frequency range 0-1000 Hz. The second level of decomposition corresponds to the frequencies 0-500 Hz (approximation A2). On the third level of decomposition the representation of the signal in the frequencies range 0-250 Hz was obtained (approximation A3).

After the decomposition parameters RA1, RA2 and RA3, expressing the root mean square values of each approximation A1, A2 and A3, were determined. Parameters RA1, RA2 and RA3 were normalized by dividing by the amplitude of the EMG signalrecorded during maximal activation of muscle and by multiplying by 100. Normalized parameters was named as nRA1, nRA2 and nRA3.

4.Statistical analysis

In order to determine the effect of muscle fatigue on the EMG signal characteristics, the statistical analysis using t-test was carried out. The independent variable was fatigue (fragments B and E), while the dependent variables were EMG signal parameters. The differences in the values of the EMG signal parameters, between B and E fragments were investigated. Statistically significant differences (p ≤ 0.05) in the values of specified parameter between the B and E fragments indicated a sensitivity of this parameter to muscle fatigue. In the analysis of differences between B and E fragments all three levels of load (50%, 60% and 70%MVC) together as well as each of the levels of load separately were included. Statistica 9.0 was used.

RESULTS

Table 1 presents the results of statistical analysis aimed at determining the impact of fatigue (the differences between fragments B and E) on the values of the parameters of the EMG signal. Mean values and 95% confidence intervals for the analysed parameters (MPF, CMPF, nRA1, nRA2 and nRA3) from triceps brachii (TB) at 3 levels of load (50, 60 and 70%MVC), at the beginning (B) and at the end (E) of the load are presented in Figure 2.

Table 1. The effect of the fatigue (fragments B and E) on the values of the EMG signal parameters from TB

(triceps brachii) muscle, obtained by t-test application

Analysis of the differences between fragments B and E, including all three levels of load (50, 60 and 70%MVC) indicates that the values of all parameters differ significantly between fragments B and E. However, in the case of the analysis for each load level separately, sensitivity to muscle fatigue is not so clear for the studied parameters. Determined on the basis of Fourier transform mean power frequency (MPF) differs significantly between fragments B and E for the load levels 60% and 70%MVC, while for 50%MVC does not show such a relationship. While the mean power frequency determined on the basis of the wavelet transform (CMPF) showed statistically significant differences between fragments B and E for each load level analysed separately. Parameters nRA2 and nRA1 are sensitive to muscle fatigue only at load level 60%MVC, whereas the parameter nRA3 at load levels 50% and 60%MVC.

Fig 2. Mean values and 95% confidence intervals for the analysed parameters (MPF, CMPF, nRA1, nRA2 and nRA3) from triceps brachii (TB) at 3 levels of load (50, 60 and 70%MVC), at the beginning (B) and at the end (E) of the load

DISCUSSION

Results of the research allowed to determine the effect of muscle fatigue on the values of the EMG signal parameters, calculated on the basis of frequency and time-frequency analysis, for different load levels of TB muscle.

Mean power frequency determined on the basis of wavelet transform (CMPF) showed statistically significant differences between fragments B and E for each load level analysed separately. While the mean power frequency determined on the basis of

Fourier transform (MPF) differs significantly between fragments B and E only for the load levels 60% and 70%MVC. This means that the mean power frequency determinedon the basis of the wavelet transform shows greater sensitivity to fatigue of muscle TB than the mean power frequency determined on the basis of Fourier transform. The obtained relationships are consistent with Larivière et al. (2008), which indicate that the indicators determined on the basis of time-frequency analysis seem to be more reliable than the indicators determined on the basis of frequency analysis. Also Barandun et al. (2009) suggests that the mean power frequency determined on the basis of both Fourier and wavelet transform can be included in the analysis of the EMG signal, but the wavelet analysis may be more suitable tool. On the other hand da Silva et al. (2008) indicate that both types of indicators show similar effectiveness in the evaluation of muscle fatigue.

The analysis of the results provided information pointing the effectiveness of the indicators of muscle fatigue calculated on the basis of the wavelet transform. Theresults indicate that the parameters determined on the basis of wavelet transform aremore effective in assessing muscle fatigue than the parameters based on the Fourier transform. These results are especially important due to the fact that the parametersof wavelet analysis do not require the adoption of simplifying assumptions about the stationarity of analysed signal.

In the analysis high levels of load (50, 60 and 70%MVC) were considered. It would be interesting to examine the effectiveness of analysed indicators of fatigue also at lowlevels of load, not exceeding 30%MVC.

CONCLUSIONS

The results will allow to more accurate assessment of muscle fatigue during work. The use of parameters describing muscle fatigue, calculated on the basis of wavelet transform allows the analysis of the EMG signal without simplifying assumptions about the stationarity of the signal.

ACKNOWLEDGEMENT

This paper has been prepared on the basis of the results of a research task carried out within the scope of the second stage of the National Programme “Improvement of safety and working conditions” partly supported in 2011-2013 – within the scope of research and development – by the Ministry of Science and Higher Education. The Central Institute for Labour Protection - National Research Institute is the Programme’s main co-ordinator.

REFERENCES

• 1. Barandun M, von Tscharner V, MeuliSimmen C, Bowen  V,  Valderrabano  V. Frequency and conduction velocity analysis of the abductor pollicis brevis muscle during early  fatigue. Journal of Electromyography and Kinesiology  2009 Feb;19(1):65–74
• 2. Bartuzi P., Tokarski T, RomanLiu D. The effect of the fatty tissue on EMG signal in young women. Acta of Bioengineering and Biomechanics 2010;12(2):87–92
• 3. Bilodeau  M,  SchindlerIvens  S,  Williams  DM,  Chandran  R,  Sharma  SS.  EMG frequency content changes with increasing force and during fatigue in the quadriceps femoris muscle of men and women. Journal of Electromyography and Kinesiology 2003 Feb;13(1):83–92
• 4. Boonstra TW, Daffertshofer A, van Ditshuizen JC, van den Heuvel MR, Hofman C, Willigenburg NW, Beek PJ. Fatiguerelated changes in motorunit synchronization of quadriceps muscles within and across legs. Journal of Electromyography and Kinesiology 2008 Oct;18(5):717–31
• 5. Da Silva RA, Larivière C, Arsenault AB, Nadeau S, Plamondon A. The comparison of wavelet and Fourierbased electromyographic indices of back muscle fatigue during dynamic contractions: validity and reliability results. Electromyography and clinical neurophysiology 2008, 48(34): 14762
• 6. Daubechies I. Ten Lectures on Wavelets. SIAM. Philadelphia. 1992
• 7. De Luca CJ. Surface electromyography: detection and recording. Boston: DelSys, 2002.
• 8. De Luca CJ. The use of surface electromyography in biomechanics. Journal of Applied Biomechanics 1997;13:135–63
• 9. Dimitrova NA, Arabadzhiev TI, Hogrel JY, Dimitrov GV. Fatigue analysis of interference EMG signals obtained from biceps brachii during isometric voluntary contraction at various force levels. Journal of Electromyography and Kinesiology 2009 Apr;19(2):252–8
• 10. Farina D, Merletti R, Indino B, Nazzaro M, Pozzo M. Surface EMG crosstalk between knee extensor muscles: experimental and model results. Muscle & Nerve 2002 Nov;26(5):681–95
• 11. Farina D, Mesin L. Sensitivity of surface EMGbased conduction velocity estimates to local tissue inhomogeneitiesinfluence of the number of channels and inter channel distance. Journal of Neuroscience Methods 2005 Mar;142(1):83–9
• 12. Flanders M. Choosing a wavelet for singletrial EMG. Journal of Neuroscience Methods 2002, 116(2): 16577
• 13. Gates DH, Dingwell JB. The effects of neuromuscular fatigue on task performance during repetitive goaldirected movements. Experimental Brain Research 2008 Jun;187(4):573–85
• 14. Gerdle B, Karlsson S, Crenshaw AG, Elert J, Fridén J. The influences of muscle fibre proportions and areas upon EMG during maximal dynamic knee extensions. European Journal of Applied Physiology 2000 Jan;81(1–2):2–10
• 15. Hermens    HJ,    Freriks   B,   DisselhorstKlug    C,   Rau    G.    Development    of recommendations for SEMG sensors and sensor placement procedures. Journal of Electromyography and Kinesiology 2000 Oct;10(5):361–74
• 16. Hermens HJ, Freriks B, Merletti R, Hagg G, Stegeman D, Blok J et al, editors. SENIAM 8: European recommendations for surface electromyography, ISBN: 90 75452152: Roessingh Research and Development bv; 1999
• 17. Hostens I, Ramon H. Assessment of muscle fatigue in low level monotonous task performance during car driving. Journal of Electromyography and Kinesiology 2005, 15: 266–74
• 18. Hostens I, Seghers J, Spaepen A, Ramon H. Validation of the wavelet spectral estimation technique in Biceps Brachii and Brachioradialis fatigue assessment during   prolonged   lowlevel   static   and   dynamic   contractions.   Journal   of Electromyography and Kinesiology 2004, 14: 205–15
• 19. Ioi H, Kawakatsu M, Nakata S, Nakasima A, Counts AL. Mechanomyogram and electromyogram analyses during isometric contraction in human masseter muscle. Australian Orthodontic Journal 2008, 24(2): 11620
• 20. KlaverKról EG, Henriquez NR, Oosterloo SJ, Klaver P, Kuipers H, Zwarts MJ. Distribution of motor unit potential velocities in the biceps brachii muscle of sprinters and endurance athletes during prolonged dynamic exercises at low force levels. Journal of Electromyography and Kinesiology 2010 Dec;20(6):1115–24
• 21. Larivière C, Gagnon D, Gravel D, Arsenault AB. The assessment of back muscle capacity using intermittent static contractions. Part I – Validity and reliability of electromyographic indices of fatigue. Journal of Electromyography and Kinesiology 2008, 18(6): 100619
• 22. Larsson B, Kadi F, Lindvall B, Gerdle B. Surface electromyography and peak torque of repetitive maximum isokinetic plantar flexions in relation to aspects of muscle morphology. Journal of Electromyography and Kinesiology 2006 Jun;16(3):281–90
• 23. Perotto AO. Anatomic guide for the electromyographer – the limbs and trunk. 4th ed. Springfield, Illinois: Charles C Thomas, 2005
• 24. Petrofsky J. The effect of the subcutaneous fat on the transfer of current through skin and into muscle. Medical Engineering & Physics 2008 Nov;30(9):1168–76
• 25. Pincivero DM, Green RC, Mark JD, Campy RM. Gender and muscle differences in EMG amplitude and median frequency, and variability during maximal voluntary contractions of the quadriceps femoris. Journal of Electromyography and Kinesiology 2000 Jun;10(3):189–96
• 26. Piscione J, Gamet D. Effect of mechanical compression due to load carrying on shoulder   muscle   fatigue   during   sustained   isometric   arm   abduction:   an electromyographic study. European Journal of Applied Physiology 2006, 97: 573 81
• 27. Pope MH, Aleksiev A, Panagiotacopulos ND, Lee JS, Wilder DG, Friesen K, Stielau W, Goel VK. Evaluation of low back muscle surface EMG signals using wavelets. Clinical Biomechanics 2000, 15(8): 56773
• 28. RomanLiu D, Konarska M. Characteristics of power spectrum density function of EMG during muscle contraction below 30% MVC. Journal of Electromyography and Kinesiology 2009 Oct, 19(5):864–74
• 29. RomanLiu D, Tokarski T, Wójcik K. Quantitative assessment of upper limb muscle fatigue depending on the conditions of repetitive task load. Journal of Electromyography and Kinesiology 2004 Dec; 14(6):671–82
• 30. Roy SH, De Luca G, Cheng MS, Johansson A, Gilmore LD, De Luca CJ. Electro mechanical stability of surface EMG sensors. Medical & Biological Engineering & Computing 2007 May;45(5):447–57
• 31. So R, Chan KM, Siu O. EMG power frequency spectrum shifts during repeated isokinetic knee and arm movements. Research Quarterly for Exercise and Sport 2002 Mar;73(1):98–106
• 32. Troiano A, Naddeo F, Sosso E, Camarota G, Merletti R, Mesin L. Assessment of force and fatigue in isometric contractions of the upper trapezius muscle by surface EMG signal and perceived exertion scale. Gait Posture. 2008, 28(2): 179 86
• 33. Visser B, van Dieën J. Pathophysiology of upper extremity muscle disorders. Journal of Electromyography and Kinesiology 2006, 16(1): 116
• 34. Von Tscharner V. Intensity analysis in timefrequency space of surface myoelectric signals  by  wavelets  of  specified  resolution.  Journal  of  Electromyography  and Kinesiology 2000, 10(6): 43345
• 35. Welch PD. The use of fast Fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Transactions on Audio Electroacoustics, Vol. AU15; 1967. pp. 70–3
• 36. Yamada E, Kusaka T, Arima N, Isobe K, Yamamoto T, Itoh S. Relationship between muscle oxygenation and electromyography  activity during sustained isometric contraction. Clinical Physiology and Functional Imaging 2008 Jul;28(4):216–21