JEPonline
Journal of 
Exercise Physiologyonline

ISSN 1097-9751

An International Electronic
Journal for Exercise Physiologists

Vol 1 No 3 October 1998
Systems Physiology: Cardiopulmonary
Reliability of electromyographic fatigue curves

JOSEPH P. WEIR, BRENT R. LLOYD, ANGIE M. TUSSING MONICA S. GREEN, S. JUANITA ROBEL

Program in Physical Therapy, University of Osteopathic Medicine and Health Sciences, Des Moines, IA


JOSEPH P. WEIR, BRENT R. LLOYD, ANGIE M. TUSSING MONICA S. GREEN, S. JUANITA ROBEL. Reliability of electromyographic fatigue curves. JEPonline Vol. 1 No. 3 1998. The analysis of electromyographic fatigue curves allows for research of muscle fatigue in vivo and has potential clinical usefulness. In general, during a fatiguing constant force sub-maximal isometric contraction, EMG amplitude increases and the frequency characteristics decrease. The purpose of this investigation was to assess the test-retest reliability of EMG fatigue curves. To this end, 21 subjects (3 male and 18 female, mean age ± SD = 33.1 ± 12.7 years) performed one minute isometric contractions of the knee extensors at 50% of maximal force on three separate days. All testing was performed at an angle 30 degrees from full extension. Each contraction was divided into 60 one-second segments. From each segment, the integrated EMG (iEMG) and median power frequency (MPF) were determined. Fatigue curves for each subject were generated by regressing the iEMG and MPF values against time. The EMG data from the maximal contraction, mean value during the fatiguing contraction, and the y-intercept of the individual regressions, were used to separately normalize the responses. The resulting slopes were analyzed with one-way repeated measures ANOVAs and by calculating intraclass correlation coefficients and standard errors of measurement (SEM). None of the ANOVAs were statistically significant (p>0.32), indicating that there were no significant mean differences across trials for any of the dependent variables. The ICC results for the iEMG data were 0.82, 0.81, and 0.77 for slopes normalized to the maximal value, the mean value, and the y-intercept, respectively. These values were not significantly different (p>0.20) from each other. The SEM values ranged from 21.9 to 33.3% of the mean values. For the MPF slopes, ICC = 0.52, 0.58, and 0.58 and the SEM values ranged from 62.0 to 69.2% of the mean values. The ICC for the MPF data normailized to the maximal response was significantly lower (p<.05) than those normalized to the mean and y-intercept, althought the differences were quite small (.06 ICC units). The iEMG ICC values were significantly (p<.05) larger then the MPF values, with the exception of comparisons between the iEMG y-intercept ICC and the MPF mean and y-intercept values (p=.069). These results indicate that higher reliability occurs with EMG fatigue curves derived from amplitude versus frequency data while normalization appears to have little effect. The relatively poor reliability for the MPF data may be due to fluctuations in electrode orientation relative to muscle fiber direction across test sessions.

Key Words: EMG, ISOMETRIC, MUSCLE 

Introduction
Muscle fatigue is an important factor in exercise performance and can affect function in clinical situations. Fatigue is often assessed by monitoring changes in muscle force production over time, however the extent of effort by the subject can be a confounding factor in the analysis. In contrast, electromyography (EMG) can be used to monitor the rate of fatigue and has the advantage of being unaffected by psychological factors such as motivation (6). As a muscle fatigues, changes in EMG amplitude and frequency characteristics occur which can be used to quantify the rate of fatigue. In general, during fatiguing submaximal constant force isometric contractions, the EMG amplitude increases and indices of the EMG frequency characteristics decrease. The change in frequency characteristics, typically assessed by monitoring the decrease in either the mean or median power frequency over time, is believed to result from a decrease in muscle fiber conduction velocity (8,17). The increase in amplitude over time has been proposed to result from an increase in motor unit recruitment and/or firing rate (6,8,11). However, the changes in amplitude have also been suggested to result from changes in muscle fiber conduction velocity (6). Regardless of the mechanistic underpinnings, the use of these parameters has potential for monitoring changes in fatigue characteristics following exercise interventions or periods of inactivity. However, the utility of this approach depends on the reliability of the measurements.

Reliability of absolute amplitude (21,22) and frequency values (4,5) of EMG signals have been studied in a variety of muscles and in general both characteristics have been shown to be reliable across test days. However, relatively little research has been performed to assess the reliability of the fatigue responses. Those studies that have been done have primarily dealt with the low back muscles and focused on the spectral characteristics (3,15,16,18). Less research has been performed examining the reliability of EMG fatigue responses using time domain analysis (7). Similarly, other muscles such as the quadriceps need to be examined since this muscle group has been shown to be critically important for normal functioning and in activities of daily living (10).

In addition, because of the large inter-individual differences in EMG amplitude and spectral characteristics, comparisons between subjects are facilitated by normalization of the EMG slope values. A variety of normalization approaches can be employed to normalize EMG fatigue curves. One approach is to normalize the slope values by the respective initial value (18), which can be estimated from the y-intercept of the individual regression analysis (12,15). Alternatively, normalization to the values recorded during a maximal voluntary contraction (MVC) have been employed (20). Finally, another normalization approach would be to normalize the fatigue slopes to the mean response during the fatiguing work bout, which would provide an index of the change in the parameter relative the average of the parameter over the trial. With the evolution of computer technology, systems that can analyze EMG signals for fatigue related changes in both the time and frequency domain may become common in clinical and research settings. However, the lack of reliability data needs to be addressed before more widespread use is warranted. Therefore, the purpose of this investigation was to assess the reliability of electromyographic fatigue curves of the quadriceps, derived from both time and frequency domain analyses, using different normalization approaches. Methods

Subjects
Twenty one subjects (three males and 18 females; age ± SD = 33.1 ± 12.7 years) volunteered to be participants in this investigation. The subjects were given a health history questionnaire and screened for exclusion criteria, such as knee pathology. Verbal and signed informed consent was obtained from each subject and the procedures were approved by the local Institutional Review Board.

Instrumentation
All testing was performed on a KinCom isokinetic dynamometer (Chattecx Corporation, Chattanooga Tennessee). Subjects were seated on the dynamometer with the hips positioned at approximately 90 degrees and with the knee positioned at 30 degrees from full extension. The axis of rotation of the dynamometer was carefully aligned with the axis of rotation of the knee joint. The subjects were stabilized with a lap belt and were additionally required to hold onto towel handles at the sides of the dynamometer chair. All positioning information was recorded for repeated use during the subsequent test days.

The EMG data was collected with an active surface EMG pre-amplifier electrode assembly (bipolar silver/silver chloride electrodes, center-to-center inter-electrode spacing = 2 cm, input impedance > 25M* at DC and > 15 M* at 100Hz; Model D-100, Therapeutics Unlimited, Iowa City IA) with a downstream driver amplifier (common mode rejection ratio = 87 dB at 60 Hz, bandwidth = 20-4000 Hz; Model EMG-554, Therapeutics Unlimited, Iowa City IA). The electrode assembly was placed over the vastus lateralis as follows. A thigh girth measurement was taken at a point 50% of the distance between the inguinal ligament and the superior pole of the patella. The electrode was placed lateral to an imaginary line bisecting the anterior superior iliac spine and the superior patellar pole at the 50% position described above. The lateral distance equaled 15% of the thigh girth measurement. This measurement protocol was repeated for each subject at each test session. We chose not to mark the skin as in previous work (21) for the convenience of the subjects and because this would facilitate subject compliance for long-term intervention studies. Following the measurements, the skin was vigorously cleaned with alcohol prior to electrode placement and the electrode assembly was secured to the muscle belly with adhesive tape. A reference electrode was placed over the crest of the tibia.

Both the EMG signal and the force signal from the dynamometer were interfaced with a BNC connector board (BNC 2080, National Instruments, Austin TX) to a 12 bit analog-to-digital converter (AT-MIO-16E-10; National Instruments, Austin TX) with a sampling rate for each channel of 1000Hz. The total system gain for the EMG signal was adjusted for each subject to allow maximum amplification without saturation of the analog-to-digital converter.

Procedures
Each subject reported to the laboratory three times for testing over a two week period with a minimum of 48 hours between each test. Following placement of the EMG electrode, the subjects performed a warm-up consisting of five minutes of stationary cycling at 150 kgm/min followed by static stretching of the anterior thigh muscles. Following positioning and stabilization on the dynamometer, the subjects performed three isometric practice trials at 50-75% of maximum followed by three maximal isometric contractions. Each maximal contraction was approximately five seconds in duration and two minutes of rest was provided between each contraction. During the contractions, the subjects were verbally encouraged to produce as much force as possible. Following the last maximal contraction, the subjects were given an additional two minutes of rest during which time the force data was analyzed to determine the MVC, which was derived from the one-second segment with the largest mean force from each of the three maximal contractions. The EMG signal from this time period was also isolated for subsequent time and frequency domain analysis (see below). The 50% MVC value was used for the fatiguing submaximal contraction and was held for just over 60 seconds. The 50% criterion was maintained by providing digital visual feedback to the subject from the computer monitor of the dynamometer. During the fatigue interval, the mean coefficient of variation in force was 3.8%. For test sessions two and three, the MVC trials were repeated in order to ensure consistency across the test sessions, however the force value used on the first test session was also used for the subsequent test sessions.

Signal Processing
All signal processing was performed using custom programs written with LabVIEW programming software (version 3.1; National Instruments, Austin TX) and details of our general approach have been reported previously (20). Briefly, the EMG signal from the fatiguing contraction was divided into 60 one-second (1-sec) segments. After low pass filtering (300 Hz cutoff, zero lag fourth order Butterworth filter), the data from each segment were analyzed in both the time and frequency domain. The time domain analysis (amplitude) consisted of digital full-wave rectification and integration of each 1-sec segment. The maximal integrated EMG (iEMG) value for each segment was used in subsequent analyses (see below). Similarly, for the frequency domain analysis, a discrete Fourier transform with a Hamming window function was used to generate a power spectral density function (PSDF) for each 1-sec segment, from which the median power frequency (MPF) was calculated. The program determined the MPF by integrating the area under the PSDF and calculating the frequency, in Hz, which divided the area of the PSDF into two equal halves. For both the amplitude and frequency data, the iEMG and MPF values were regressed against time and the resulting slopes used as dependent variables in the statistical calculations.

Statistical Analyses
For each subject, the slopes of the iEMG and MPF versus time relationships were separately normalized using three procedures: normalization with the respective value from the MVC, normalization with the mean EMG response during the fatiguing task, and normalization with the y-intercepts of the individual regressions. For each of the six procedures, the group data were examined by calculating a one-way repeated measures ANOVA [with the Huynh-Feldt correction; effect sizes (w2) were calculated as described by Keppel (11)], a two-way intraclass correlation coefficient (ICC) as described by Baumgartner (2), and a standard error of measurement (SEM). Differences between ICC values determined with different normalization procedures were examined using the procedure described by Alsawalmeh and Feldt (1). In addition, 95% confidence intervals for the ICC values were calculated as described by Morrow and Jackson (14). Alpha levels of 0.05 were used for statistical significance testing.

Results
The mean values across trials are presented in  Figure 1 and Figure 2. For all ANOVAs, the trial effect was not significant (p> 0.32; effect sizes (w2) ranged from -0.009 to 0.003) which indicates that there were no systematic differences in EMG slopes across trials. The ICC and associated SEM values for the iEMG and MPF data are presented in Table 1. In general, the ICC values were higher for the iEMG data than for the MPF data. Similarly, the SEM values, when expressed as a percentage of the mean, were lower for the iEMG than for the MPF data. For the iEMG data, the ICCs were not significantly different (p>0.20) between normalization procedures. The MPF ICCs for the data normalized to MVC was significantly lower (p<.05) than those from the mean and y-intercept normalizations. The iEMG ICC values were significantly larger (p<.05) than the MPF values with the exception of the comparisons between the iEMG y-intercept ICC and the MPF mean and y-intercept values (p=.069).

Discussion
The results of the ANOVA analyses indicated no significant changes in mean responses across test session for any of the six dependent variables. These results are unlikely to be due to Type II errors as the effect size calculations (w2) ranged from -0.009 (negative w2 values reflect F ratios less than 1.0) to 0.0003, indicating a trivially small effect. However, since this analysis only indicates systematic changes across test sessions, these results show that there was no effect of the repeated testing per se on the fatigue curves. Significant effects would only be apparent if most or all subjects exhibited the same pattern of change across tests. In contrast, between subjects variability in the pattern of scores across trials results in an increase in the error term of the ANOVA analysis. However this source of variability, which will decrease the power of the ANOVA, is a source of variability that will depress the ICC and inflate the SEM data described below.

The ICC and SEM values reported in Table 1 indicate that while the reliability for the iEMG fatigue data was relatively good, the MPF fatigue data exhibited relatively poor reliability. The iEMG ICC values normalized for the MVC and the mean were significantly larger than all the MPF values, while the iEMG y-intercept value was significantly larger than the MPF MVC value and approached significance (p=.069) for the y-intercept and mean normalizations. The poor reliability for the MPF results are in contrast to previous work with the low back muscles reporting relatively good reliability for spectral fatigue data (15,16,18).

It may be argued that variability across trials in the MPF data may have been introduced by errors in electrode placement. Indeed, as noted previously we chose not to mark the skin to help in consistent electrode placement. Instead, we used a measurement protocol that in our pilot work allowed us to place the electrode over the vastus lateralis. However, this effect should also have reduced the reliability the iEMG data, yet the reliability coefficients for those analyses were notably higher. We feel the likely explanation for the lower reliability of the MPF data than the iEMG data (as well as lower reliability coefficients than the low back studies cited above) is the possible effect of the orientation of the electrode relative to the muscle fiber direction. That is, variations in the angle between the fiber direction and a line connecting the centers of the pick up sites of the electrode on repeated testing may have led to the reliability results reported here. Because changes in MPF during fatigue are purported to reflect changes in muscle fiber conduction velocity, alterations in electrode orientation relative to fiber direction are more likely to affect MPF fatigue indices than iEMG measures. Consistent with this hypothesis, we have found that alterations in electrode orientation can drastically reduce the correlation coefficients between MPF derived fatigue curves collected from the same muscle at the same time while the effect of alteration of electrode orientation on iEMG fatigue curves was much less than for the MPF data (19).

In contrast to the MPF data, the iEMG fatigue curves (irrespective of normalization procedure) showed relatively high reliability coefficients. These results are consistent with those of deVries (7) who used a similar approach with the forearm flexors. These results suggest that time domain derived fatigue curves may be a useful tool for assessing fatigue responses in the leg extensors. In addition, the iEMG data, in conjunction with the MPF data, indicate that assessment of fatigue in the leg extensors from interpretation of changes over time in spectral fatigue curves should be done with caution. Future research needs to be performed to enhance the reliability of MPF fatigue curves in the leg extensors.

The effect of normalization procedure within the iEMG and MPF analyses was not large. The differences in the iEMG data were within .05 ICC units and there were no significant differences in ICC values between iEMG normalization procedures. For the MPF data, the mean and y-intercept normalization procedures resulted in equal ICC values (0.58), while the MVC normalization was significantly lower than both. From a practical perspective, however, the differences between normalization procedures was small (.06 ICC units) and since all MPF normalization procedures showed relatively poor reliability, the small difference between the normalization to MVC and the other procedures has little consequence.

In summary, the primary finding of this investigation was that while the reliability of the iEMG fatigue data was relatively good, the MPF fatigue data exhibited poor reliability. In both cases, the normalization procedure appears to have little effect. The lower reliability associated with the MPF data may be due to small fluctuations in the orientation of the electrodes upon repeated testing over the different test days. Regardless of the cause of lower reliability of the MPF data, it appears that routine use of such measures for the quadriceps requires procedures to enhance reliability. In contrast, time domain measures, regardless of normalization procedure, have good reliability and are therefore more attractive at this time. 

References
1. Alsawalmeh YM, Feldt LS. Testing the equality of two related intraclass reliability coefficients. Appl Psychol Meas 1994, 18: 183-190.
2. Baumgartner TA. Norm-referenced measurement: Reliability. In: Safrit MJ, Wood TM, editors. Measurement Concepts in Physical Education and Exercise Science, Champaign, Il: Human Kinetics, 1989: 45-72.
3. Biederman HJ, Shanks GL, Inglis J. Median frequency estimates of paraspinal muscles: reliability analysis. Electromyogr Clin Neurophysiol 1990; 30: 83-88.
4. Bilodeau M, Arsenault AB, Gravel D, Bourbonnais D. EMG power spectrum of elbow extensors: A reliability study. Electromyogr Clin Neurophysiol 1994; 34: 149-158.
5. Daanen HAM, Mazure M, Holewijn M, Van der Velde EA. Reproducibility of the mean power frequency of the surface electromyogram. Eur J Appl Physio 1990; 61: 274-277.
6. DeLuca CJ. Myoelectrical manifestations of localized muscular fatigue in humans. CRC Crit Rev Biomed Engr 1984; 11: 251-279.
7. deVries HA. Method for evaluation of muscle fatigue and endurance from electromyographic fatigue curves. Am J Phys Med Rehabil 1968; 47: 125-135.
8. Eberstein A, Beattie B. Simultaneous measurement of muscle conduction velocity and EMG power spectrum changes during fatigue. Muscle Nerve 1985; 8: 768-773.
9. Edwards RG, Lippold OCJ. The relation between force and integrated electrical activity in fatigued muscles. J Physiol (Lond) 1956; 132: 677-681.
10. Fiatarone MA, Marks EC, Ryan ND, Meredith CN, Lipsitz LA, Evans WJ. High-intensity strength training in nonagenarians. Effects on skeletal muscle. J Am Med Assoc 1990; 263: 3029-3034.
11. Keppel G. Design and Analysis. A Researcher's Handbook (3rd Ed). Englewood Cliffs, NJ: Prentice Hall, 1991: 354-355. 12. Mannion AF, Dolan P. Relationship between myoelectric and mechanical manifestations of fatigue in the quadriceps femoris muscle group. Eur J Appl Physiol 1996; 74: 411-419.
12. Mannion AF, Dolan P. Relationship between myoelectric and mechanical manifestations of fatigue in the quadriceps femoris muscle group. Eur J Appl Physiol, 1996; 74: 411- 419.
13. Moritani T., Muro M, Nagata A. Intramuscular and surface electromyogram changes during muscle fatigue. J Appl Physiol 1986; 60: 1179-1185.
14. Morrow JR, Jackson AW. How "significant" is your reliability? Res Quart Exerc Sport 1993; 64: 352-355.
15. Ng JKF, Richardson CA. Reliability of electromyographic power spectral analysis of back muscle endurance in healthy subjects. Arch Phys Med Rehabil 1996; 77: 259-264.
16. Oliver CW, Tillotson KM, Jones APC, Royal RA, Greenough CG. Reproducibility of lumbar paraspinal surface electromyogram power spectra. Clin Biomech 1996; 11: 317-321.
17. Sadoyama T, Masuda T, Miyano H. Relationships between muscle fibre conduction velocity and frequency parameters of surface EMG during sustained contraction. Eur J Appl Physiol 1983; 51: 247-256.
18. van Dieen JH, Heijblom P. Reproducibility of isometric trunk extension torque, trunk extensor endurance, and related electromyographic parameters in the context of their clinical applicability. J Orthop Res 1996; 14: 139-143.
19. Weir JP, Davis A, Mahoney K, Haan K. Assessing fatigue with electromyography: Influence of electrode orientation. J Strength Cond Res 1997; 11: 288 (abstract).
20. Weir JP, McDonough AL, Hill VJ. The effects of joint angle on electromyographic indices of fatigue. Eur J Appl Physiol 1996; 73: 387-392.
21. Weir JP, Wagner LL, Housh TJ. Linearity and reliability of the IEMG V torque relationship for the forearm flexors and leg extensors. Am J Phys Med Rehabil 1992; 71: 283-287.
22. Yang JF, Winter DA. Electromyography reliability in maximal and submaximal isometric contractions. Arch Phys Med Rehabil 1983; 64: 417-420. 

Address all correspondence to: Joseph P. Weir, Program In Physical Therapy, University of Osteopathic Medicine and Health Sciences, 3200 Grand Avenue, Des Moines, IA 50312 (515) 271-1733 (515) 271-1714 (Fax)
Copyright ©1998
American Society of Exercise Physiologists
All Rights Reserved


ASEP Table of Contents