Normalisation of gait EMGs: a re-examination

https://doi.org/10.1016/S1050-6411(03)00082-8Get rights and content

Abstract

The purpose of this study was to compare four different methods of normalising electromyograms (EMGs) recorded during normal gait. Comparisons were made between the amplitude, intra-individual variability and inter-individual variability of EMGs. Surface EMGs were recorded from the biceps femoris, semitendinosus, vastus lateralis and vastus medialis of ten males and two females while they walked on a treadmill at a self-selected speed. EMGs from the same muscles were subsequently recorded during isometric maximal voluntary contractions (MVCs) and concentric, isokinetic MVCs that were performed between 0.52 and 7.85 rad·s−1 on a BIODEX dynamometer. EMGs were also recorded during eccentric, isokinetic MVCs between 0.52 and 2.62 rad·s−1. Gait EMGs were then normalised at 2% intervals of the gait cycle by expressing them as a percentage of the following reference values: the mean (mean dynamic method) and the peak (peak dynamic method) EMG from the intra-individual ensemble average; the EMG from an isometric MVC (isometric MVC method); and the EMG from an isokinetic MVC that occurred with the same muscle action, length and velocity of musculotendinous unit as the gait EMGs (isokinetic MVC method). The isokinetic MVC method produced significantly greater (P<0.05) intra-individual variability compared to the other methods when it was measured using the variance ratio. Inter-individual variability of gait EMGs, again measured using the variance ratio, was also greatest when they were normalised using the isokinetic MVC method. The pattern and amplitude of EMGs normalised using the isometric MVC method and the isokinetic MVC method were very similar (root mean square difference and absolute difference both less than 3%). It was concluded that the isokinetic MVC method should not be adopted by gait researchers or clinicians as it does not reduce intra- or inter-individual variability anymore than existing normalisation methods, nor does it provide a more representative measure of muscle activation during gait than the isometric MVC method.

Introduction

The importance of normalising electromyograms (EMGs) has long been recognised by researchers and clinicians who record them during gait analysis (e.g. [27], [29], [35], [38]). Gait EMGs were first normalised [14] using a method that divided each point that constitutes the processed EMG by the peak value recorded from the same EMG. This method, subsequently referred to as the peak dynamic method, still appears to be popular among gait electromyographers (e.g. [19], [31]). A second and equally popular method, introduced by Dubo et al. [13], divided each data point included in the gait EMG by the peak EMG from an isometric maximal voluntary contraction (MVC) of the same muscle, which is usually performed in the middle of the range of motion. This method, subsequently referred to as the isometric MVC method, was later adopted by Arsenault et al. (e.g. [4]).

Yang and Winter [41] compared a number of normalisation methods in an attempt to establish which would provide a normal gait EMG template and, therefore, improve the use of electromyography as a diagnostic tool in gait analysis. Based on this rationale, the criterion for selecting the best method was the one that most reduced the inter-individual variability of ensemble averaged EMGs [41]. The peak dynamic method and the mean dynamic method (a similar but previously unpublished method that divided each data point within the gait EMG by the mean value recorded from the same EMG) were included in the comparison. Yang and Winter [41] noted that these two methods do not have the potential to provide any information on the degree of muscle activation that occurs during gait. The authors chose not to include the isometric MVC method in their comparison as they had previously [40] reported that EMGs from isometric MVCs displayed poor reliability. Instead, they also included a method that used the peak EMG from an isometric sub-maximal (50%) voluntary contraction of the same muscle as the denominator in the normalisation equation subsequently referred to as the sub-MVC method. Yang and Winter [41] discovered that the peak dynamic method and the mean dynamic method both reduced inter-individual variability in comparison to the un-normalised EMGs and those normalised by the sub-MVC method. As a consequence of this study, the mean dynamic method was adopted by Winter’s group (e.g. [39]) and a number of other researchers (e.g. [25], [28]). Further evidence that the mean dynamic method reduces inter-individual variability more than other normalisation methods was provided by Shiavi et al. [34], [35] who also compared the mean dynamic method with the peak dynamic method. They discovered that the mean dynamic method was slightly better at reducing inter-individual variability, particularly during periods of muscle quiescence, due to a relatively lower standard deviation during these periods. Based on this evidence, Shiavi et al. [34], [35] also advocated the use of the mean dynamic method particularly when wishing to distinguish between periods of muscle activity and inactivity during gait. However, more recently, Allison et al. [2] and Knutson et al. [23] both warned against using the mean dynamic method or peak dynamic method, although not specifically in gait analysis, as they may remove the true biological variation within a group.

By virtue of the nature of the denominator in its normalisation equation, the isometric MVC method is the only one that has the potential to reveal how active a muscle is during gait. Despite criticism by Yang and Winter [41], the isometric MVC method continues to be used in gait analysis (e.g. [7], [12]). However, despite continued use, only Dubo et al. [13] have questioned the suitability of using the EMG from an isometric MVC to normalise gait EMGs, which are clearly recorded during non-isometric contractions. Only a small number of studies have investigated the suitability of using the isometric MVC method to normalise EMGs from non-isometric tasks. Mirka [24] normalised EMGs recorded from sub-maximal isokinetic contractions of the trunk flexor and extensor muscles. Relatively large differences (15–50%) were reported between the isometric MVC method and a method that normalised the task EMGs using EMGs from isometric MVCs performed at the same trunk angle that occurred during the task. In contrast, Knudson and Johnston [22] later reported average absolute differences of less than 7% between the same two methods when used to normalise lower limb muscle EMGs recorded during standing from a seated position. Furthermore, Kellis and Baltzopoulos [21] were the first authors to use EMGs from isokinetic MVCs that had the same muscle action (i.e. concentric or eccentric), joint angle and joint angular velocity as the task EMG as the denominator in the normalisation equation. They discovered a significantly greater (P<0.05) normalised muscle activation amplitude from this method, subsequently referred to as the isokinetic MVC method, when compared to the isometric MVC method when both were used to normalise EMGs from knee flexors and extensors when acting as antagonists during isokinetic MVCs. However, biceps brachii EMGs recorded during isotonic elbow flexions and extensions were not significantly different (P=0.315) when normalised by the isometric MVC method and the isokinetic MVC method [5]. Further comparisons between the isometric MVC method and the isokinetic MVC method are warranted as some researchers have reported that peak EMGs from the knee extensors are greater during concentric MVCs than during either eccentric (e.g. [32], [37]) or isometric (e.g. [6]) MVCs. Alternatively, Ghori et al. [17] and Amiridis et al. [3] demonstrated no difference in peak EMG between different muscle actions. Furthermore, a number of researchers (e.g. [3], [6], [37]) have reported that the amplitude of EMGs from knee extensor muscles increased, rather than remaining constant, as the velocity of the isokinetic dynamometer’s lever arm increased during concentric MVCs.

Section snippets

Aims

Based on the above evidence, the general opinion provided by gait related review articles is that the peak dynamic method or, preferably, the mean dynamic method should be used if the aim of the analysis is to reduce inter-individual variability and produce a general gait-EMG template (e.g. [33], [38]). Alternatively, the isometric MVC method is recommended when the goal is to ascertain the level of muscle activity that is required to walk (e.g. [9], [10], [29]). However, these opinions are not

Participants

Ten males ((mean±SD) age 31.7±5.9 years; height 1.81±0.04 m; mass 80.0±12.2 kg) and two females (age 34.5±2.1 years; height 1.65±0.01 m; mass 61.0±1.4 kg) took part in the investigation after reading and signing an informed consent form. All participants were regularly involved in activities that exercised their quadriceps and hamstrings and had no known gait pathology.

Experimental design

To enable EMGs recorded from lower limb muscles during gait to be normalised by the isometric MVC method and the isokinetic MVC

Intra-individual ensemble averages

Gait EMGs normalised by the isometric MVC method and isokinetic MVC method are shown, for one individual, in Fig. 2, Fig. 3, respectively. The output of both of these methods revealed that gait EMGs were generally less than 20% of the amplitude recorded from MVCs. Some sections of the ensemble averages in Fig. 3 are missing as they could not all be normalised using the isokinetic MVC method. These gaps occurred because not all the maximum lengthening velocities experienced during gait were

Discussion

The overall aim of this investigation was to re-evaluate the three methods that have commonly been used to normalise EMGs recorded during normal gait, and to compare them with the isokinetic MVC method that has not previously been used in gait analysis. The outputs of both the mean dynamic and peak dynamic normalisation methods only serve to inform the researcher or clinician about the level of activity displayed by a muscle throughout the gait cycle in relation to the average and maximum

Conclusion

The isokinetic MVC method resulted in EMGs that had greater intra-individual variability, and were therefore less reliable, than un-normalised EMGs or those normalised by the mean dynamic, peak dynamic or isometric MVC methods. The findings of this investigation also agree with previous research [23], [34], [35], [41] that the dynamic methods, and in particular the mean dynamic method, yield the most homogeneous templates of muscle activity during gait. This is noteworthy as the variance ratio

Acknowledgements

The first author would like to acknowledge the assistance of David Clark in matching gait EMGs with those recorded from MVCs.

Adrian Burden received a B.Sc. degree in sports science from Crewe and Alsager College (1987), a masters from the University of Salford (1991) and a Ph.D. from the University of Brighton (2002). He has held lecturing positions in biomechanics at Brunel University and the University of Brighton, and is currently course leader for the sport, exercise and coaching science degree in the Department of Exercise and Sport Science at Manchester Metropolitan University. His main research interests are

References (41)

  • G.M. Allen et al.

    Reliability of measurements of muscle strength and voluntary activation using twitch interpolation

    Muscle and Nerve

    (1995)
  • I.G. Amiridis et al.

    Co-activation and tension regulating phenomena during isokinetic knee extension in sedentary and highly skilled humans

    European Journal of Applied Physiology

    (1996)
  • A.B. Arsenault et al.

    How many strides are required for the analysis of electromyographic data in gait?

    Scandinavian Journal of Rehabilitative Medicine

    (1986)
  • M.F. Bobbert et al.

    Evaluation of moment–angle curves in isokinetic knee extension

    Medicine and Science in Sports and Exercise

    (1992)
  • M.G. Ciccotti et al.

    An electromyographic analysis of the knee during functional activities—I. The normal profile

    The American Journal of Sports Medicine

    (1994)
  • R.B. Davis et al.

    Analysis of gait

  • R.B. Davis et al.

    Gait data: reporting, archiving and sharing

  • S.L. Delp et al.

    An interactive graphics-based model of the lower extremity to study orthopaedic surgical procedures

    IEEE Transactions on Biomedical Engineering

    (1990)
  • G.Y. Diaz et al.

    Electromyographic analysis of selected lower extremity musculature in normal subjects during ambulation with and without a ‘protonics’ knee brace

    Journal of Orthopaedic and Sports Physical Therapy

    (1997)
  • H.I.C. Dubo et al.

    Electromyographic temporal analysis of gait: normal human locomotion

    Archives of Physical Medicine and Rehabilitation

    (1976)
  • Cited by (0)

    Adrian Burden received a B.Sc. degree in sports science from Crewe and Alsager College (1987), a masters from the University of Salford (1991) and a Ph.D. from the University of Brighton (2002). He has held lecturing positions in biomechanics at Brunel University and the University of Brighton, and is currently course leader for the sport, exercise and coaching science degree in the Department of Exercise and Sport Science at Manchester Metropolitan University. His main research interests are the processing and normalisation of electromyograms, and the biomechanics of low back pain.

    Marion Trew began her career as a clinical physiotherapist working both in the UK and in Thailand. In 1976, she became a lecturer in physiotherapy at Coventry Polytechnic and in 1989 moved to the University of Brighton to set up occupational therapy and physiotherapy education. Her current research focusses on aspects of exercise and rehabilitation, and includes two projects on exercise in people of retirement age and one related to exercise and Parkinson’s disease.

    Vasilios Baltzopoulos received a B.Sc. degree in sports and exercise science from the Aristotelian University of Thessaloniki in Greece and a masters (1988) and Ph.D. degree (1991) from the University of Liverpool in England. He has held positions as lecturer in biomechanics at the University of Liverpool, reader in biomechanics at the Manchester Metropolitan University (MMU) and associate professor of biomechanics at the University of Thessaly in Greece. He is currently a professor of musculoskeletal biomechanics at the Centre for Biophysical and Clinical Research into Human Movement at MMU. His main research interests are measurement and modelling of joint and muscle function and biomechanical data modelling and processing techniques used in inverse dynamics applications.

    View full text