Electroencephalography and clinical Neurophysiology 109 (1998) 135–139
Locomotor training in paraplegic patients: a new approach to assess changes in leg muscle EMG patterns Th. Erni*, G. Colombo Paraplegic Centre, University Hospital Balgrist, Forchstrasse 340, CH-8008 Zu¨rich, Switzerland Accepted for publication: 5 December 1997
Abstract This study describes an amplitude independent assessment of changes in leg muscle EMG patterns in both complete and incomplete paraplegic patients during the course of locomotor training. The approach expresses the change as an approximation of the patients’ gait EMG pattern compared with that of healthy subjects. The variation ratio (VR), coefficient of variation (CV) and Pearson’s correlation coefficient (R), are used as measures of the dissimilarity/similarity of a set of wave forms. These parameters were evaluated for their ability to assess changes in the EMG pattern of the patients with respect to that of healthy subjects. The VR showed the best correlation to our data and was therefore considered to represent the optimum variable in the assessment of changes in EMG patterns. 1998 Elsevier Science Ireland Ltd. Keywords: Electromyogram (EMG); Gait analysis; Statistics; Locomotor training
1. Introduction Recent studies have demonstrated an improvement in locomotor function to be associated with an increase in EMG activity in leg extensor muscles in both complete and incomplete paraplegic patients (Dietz et al., 1994, 1995). The improvement was observed during the course of a daily locomotor training programme that started after a post-trauma period of 4–5 weeks (see also Dietz, 1997; Wernig et al., 1995). The EMG activity was recorded while the patients were walking on a treadmill. The assistance of physiotherapists was kept to a minimum. Signal energy was calculated (i.e. the root mean square value of the EMG (EMG RMS)) during defined periods of the stance and swing phases. Changes in leg muscle EMG activity were analysed using theses values as a function of training. It was shown that during the course of the training a significant increase in EMG amplitude occurred in the medial gastrocnemius muscle during the stance phase while there was no change during the swing phase. No clear effect was seen
* Corresponding author. Tel.: +41 3863737; fax: +41 386 3909; e-mail:
[email protected]
0924-980X/98/$19.00 1998 Elsevier Science Ireland Ltd. All rights reserved PII S0924-980X (98)0000 5-8
during the swing phase for the tibialis anterior muscle (for details see Dietz et al., 1995). The EMG RMS, as an amplitude-dependent measure, is influenced by several variables which differ not only between but also within subjects and even on a day to day basis. Such factors include for example skin resistance and the degree of body support (unloading). Therefore, statistical analysis of RMS data may not always be appropriate. Furthermore, a change in EMG RMS during the course of training provides little information regarding changes of the modulation of EMG activity. The aim of this study was to utilise amplitude-independent EMG analysis techniques in the hope that they would reveal recovery of the patients’ EMG activity (i.e. the overall EMG modulation (EM)) towards that of healthy subjects under comparable walking conditions. Three possible analysis techniques were considered namely Pearson’s correlation coefficient (R), the variation ratio (VR) (see Hershler and Milner, 1978; Kadaba et al., 1985) and the coefficient of variation (CV) (see Winter and Yack, 1987; Winter, 1991). As the intention of this study was to assess the changes in the overall EM during gait, analysis techniques that compared specific phases of the patients’ gait pattern with corresponding phases of the pattern recorded in healthy
EEM 97587
136
Th. Erni, G. Colombo / Electroencephalography and clinical Neurophysiology 109 (1998) 135–139
subjects were not considered. However, the analysis techniques used in this study would not reveal changes in the pathophysiological characteristics of the patients EM during the course of training.
2. Methods Data were collected from 6 patients (all male, mean age 34 years) with either complete or incomplete paraplegia. After a post-trauma period of 4–5 weeks they underwent a daily locomotor training on a treadmill (speed approximately 1.5 km/h). Every week EMG recordings were made using surface electrodes from the medial gastrocnemius and tibialis anterior muscles of both legs. Physiotherapists were on hand to assist leg movement but their involvement was kept to a minimum. The patients’ body weight was partially unloaded using a parachute harness connected to an overhead crane. This body weight support was necessary in order to allow stepping movements to be performed because the patients’ leg muscles were unable to support the full body weight (Barbeau and Rossignol, 1994; Dietz et al., 1994, 1995). The same recordings were taken once each from 16 healthy subjects (10 men, 6 women, mean age 31 years) who had neither body unloading nor help from the physiotherapists. 2.1. Data analysis For a detailed description of the recording techniques and signal analysis, see articles by Dietz et al. (1994, 1995). Briefly, all signals were sampled at 600 Hz and individual step cycles, triggered by the impact of the right heel, were normalised to a relative time scale of one step cycle starting and ending with the right heel strike. Each step cycle consisted of 1001 temporal points. The EMG signal was rectified and for each stride the EMG profile was normalised by setting its mean to 100% and smoothed using a 17-pointwindow-filter. To provide the window with its required full complement of points, missing points at the beginning of the stride were obtained from its end and vice versa (Yang and Winter, 1984; Winter and Yack, 1987; Gabel and Brand, 1994). The patients’ mean EMG over 16 strides of each recording session was calculated for each muscle. For each of the 16 healthy subjects the mean EMG over 100 strides was calculated and from this the grand mean for all 16 healthy subjects was calculated. To compare the EM of the patients with that of healthy subjects during gait, 3 different techniques were used. (1) Pearson’s correlation coefficient (R). This was calculated between the patients’ mean EMG profile of each session and the healthy subjects grand mean EMG profile. m
R=
¯ ¯ ∑ (Epi − Ep)(Eh i − Eh)
i=1
m ⋅ sp ⋅ sh
(1)
where m is the number of temporal points, Epi is the patients’ mean EMG at time epoch i, E¯p is the average of the patients’ mean EMG, sp is the standard deviation of the patients’ mean EMG, E¯hi is the healthy subjects’ grand mean EMG at time epoch i; Eh is the average of the healthy subjects’ grand mean EMG and sh is the standard deviation of the healthy subjects’ grand mean EMG. Higher R values are indicative of greater similarities of the EM between patients and healthy subjects. (2) The variation ratio (VR). The VR was calculated according to Hershler and Milner (1978) and Kadaba et al. (1985). m
n
∑ ∑ (Eij − E¯ 1x
7i )2 =m(n − 1)
VR =
i=1 j=1
M
n
(2)
¯ 2 =(mn − 1) ∑ ∑ (Eij − E)
i=1 j=1
where m is the number of temporal points, n is the number of gait cycles over which VR is evaluated, Eij is the value of the jth EMG signal at time epoch i, E¯i is the average of EMG values at time epoch i averaged over j gait cycles and E¯ is the grand mean average of the EMG signal. The VR provides a measure of the repeatability of a waveform over a given number of identical gait cycles. For total dissimilar wave forms E¯i → E¯ and the VR tends to one; for completely reproducible wave forms Eij → Ei and the VR tends to zero.. To calculate the VR for each recording session the 16 single stride EMG profiles of the patient and the 16 mean EMG profiles from healthy subjects were pooled. The VR was then evaluated for all 32 EMG profiles. The more similar the patients’ and the healthy subjects’ EM the smaller the VR. (3) The coefficient of variation (CV). This value was calculated according to Winter and Yack (1987) and Winter (1991).
r 1 m
CV =
∑ s2i
m i=1 1 m ¯ ∑E m i=1 i
(3)
where m is the number of temporal points, E¯i is the average of EMG values at time epoch i averaged over all gait cycles and si is the standard deviation of the EMG signal at time epoch i. The CV can be considered as a measure of the variance-to-signal ratio. The CV for each measurement session was calculated over the same 32 EMG profiles that were used to calculate the VR. Smaller CVs are indicative of similarities between waveforms. One problem that was encountered when calculating all 3 variables was that most EMG profiles for the patients were out of phase compared with the healthy subjects. The degree of this shift differed from session to session. Because VR, CV and R are influenced by these phase shifts they were corrected using the following procedure. The phase shift of the best cross-correlation (calculated with a cross-correla-
Th. Erni, G. Colombo / Electroencephalography and clinical Neurophysiology 109 (1998) 135–139
tion procedure provided by SAS 6.11) between the mean EMG profile of patients and healthy subjects was determined and then altered within a given interval until the best VR, CV or R was found. The VR, CV and R were compared using Pearson’s correlation coefficient and Pearson’s partial correlation for each pair, controlling for the effect of the third variable.
3. Results Fig. 1 shows an example of the mean EMG profile of the right medial gastrocnemius for two patients. At the beginning of the training programme neither patient could walk on the treadmill unassisted. For Patient 1 assistance was necessary throughout the course of training, while Patient 2 was able to walk unaided by the end of the training programme.
137
The mean EMG profile of Patient 1 did not change during the course of training and therefore neither VR, CV nor R values for this patient showed significant changes during the training period (Fig. 2). Initially the EM of Patient 2 was different to that of the healthy subjects but became more similar towards the end of the training (Fig. 1). This effect was reflected as a significant change of VR and CV during the course of training (Fig. 2). The correlation coefficients between the 3 variables were both high and significant (Table 1). The partial correlation revealed a significant correlation only between VR and CV and VR and R. The partial correlation between CV and R was not significant.
4. Discussion As shown in Figs. 1 and 2 all 3 tested variables, VR, CV
Fig. 1. Mean EMG profile from the right medial gastrocnemius over one stride triggered by heel strike. The amplitude was normalised to the average EMG over one stride. Patient 1 shows no improvement of leg muscle function as indicated by the EMG profiles being dissimilar before and after training compared with the healthy subject’s profile. Patient 2 demonstrated improvement of leg muscle function. This patient was able to walk unaided by the end of the training programme. The EMG profile after training was much more similar to the healthy subjects’ profile than it was at the onset of the training programme.
138
Th. Erni, G. Colombo / Electroencephalography and clinical Neurophysiology 109 (1998) 135–139
and R, were able to differentiate an improvement of leg muscle EMG pattern during training. However, only the VR and the CV detected the significance of the improvement of Patient 2. As expected the correlation analysis showed a tight relationship between VR, CV and R as both VR and CV measure the dissimilarity of the shape of a group of 32 patients’ and healthy subjects’ EMG profiles whereas R expresses the similarity of two mean profiles for the patients’ and the healthy subjects’ 16 EMG profiles. The partial correlation between VR, CV and R revealed that
the VR contains most information relating to changes in EM because the correlation between CV and R drops to nearly zero if the influence of the VR on this correlation is controlled. If only the mean EMG profiles are compared, as was done with R, much information contained in the single step profiles is lost. This is especially so for the patients. At the beginning of the training there are large differences in the EMG modulation between different strides. During the course of training their EMG profiles can become more similar to those of healthy subjects and can also become
Fig. 2. The changes of VR, CV and R during the course of locomotor training. The EMG measurement is plotted as a function of the weekly session number. r, Pearson’s correlation coefficient between VR, CV, R and session number. P, significance value.
Th. Erni, G. Colombo / Electroencephalography and clinical Neurophysiology 109 (1998) 135–139
139
Acknowledgements
Table 1 Correlation between VR, CV, R
VR-CV VR-R CV-R
r
Partial r
0.816* −0.851* −0.668*
0.633* −0.711* 0.086
This work was supported by grants from the Swiss National Science Foundation (No. 3100-042899.95) and the International Research Institute for Paraplegia (P16/ 93). We thank V. Dietz and J. Gibson for corrections and M. Stu¨ssi for technical assistance.
r: Pearson’s correlation coefficient; *P , 0.001; n = 96
References more stable between different strides. The VR and the CV take this into account. Kadaba et al. (1985) tried to compare the repeatability of gait EMG data measured with surface and intramuscular wire electrodes using the VR analysis. They reported that the VR could differentiate the signals recorded from the two types of electrodes and therefore suggested that the VR could be used as an index of the repeatability of wave forms. Gabel and Brand (1994) investigated the effects of the number of strides used and the amount of smoothing applied to the EMG data on VR and CV. They showed that, in contrast to the CV, the VR is independent of the number of strides used and is only minimally affected by the amount of smoothing applied to the data. Because of the above arguments we decided to use the VR as an amplitude-independent measure of the similarity between the patients’ and healthy subjects’ EMG profiles. It is accepted that daily locomotor training on a treadmill improves both walking ability in incomplete paraplegic patients and leg muscle function in complete paraplegic patients (Wernig et al., 1992; Dietz et al., 1994; Dietz, 1997). Using the RMS technique as an amplitude-dependent measurement it could be shown that this recovery of walking ability is accompanied by an improvement of leg muscle EMG activity (Dietz et al., 1994, 1995). Whether or not the changes in the EM as measured with the VR also reflect an improvement of leg muscle EMG activity requires further investigation.
Barbeau, H. and Rossignol, S. Enhancement of locomotor recovery following spinal cord injury. Curr. Opin. Neurol., 1994, 7: 517–524. Dietz, V. Neurophysiology of gait disorders: present and future applications. Electroenceph. clin. Neurophysiol., 1997, 103 (3): 333–355. Dietz, V., Colombo, G. and Jensen, L. Locomotor activity in spinal man. Lancet, 1994, 344 (8932): 1260–1263. Dietz, V., Colombo, G., Jensen, L. and Baumgartner, L. Locomotor capacity of spinal cord in paraplegic patients. Ann. Neurol., 1995, 37 (5): 574–582. Gabel, R.H. and Brand, A. The effects of signal conditioning on the statistical analyses of gait EMG. Electroenceph. clin. Neurophysiol., 1994, 93: 188–201. Hershler, C. and Milner, M. An optimality criterion for processing electromyographic (EMG) signals relating to human locomotion. IEEE Trans. Biomed. Eng., BME, 1978, 25 (5): 413–420. Kadaba, M.P., Wootten, M.E., Gainey, J. and Cochran, G.V.B. Repeatability of phasic muscle activity: performance of surface and intramuscular wire electrodes in gait analysis. J. Orthop. Res., 1985, 3: 350–359. Wernig, A. and Mu¨ller, S. Laufband locomotion with body weight support improved walking in persons with severe spinal cord injuries. Paraplegia, 1992, 30: 229–238. Wernig, A., Mu¨ller, S., Nanassy, A. and Cagol, E. Laufband therapy based on ‘rules of spinal locomotion’ is effective in spinal cord-injured persons. Eur. J. Neurosci., 1995, 7: 823–829. Winter, D.A.. The Biomechanics and Motor Control of Human Gait, University of Waterloo Press, Waterloo, 1991. Winter, D.A. and Yack, H.J. EMG profiles during normal human walking: stride-to-stride and inter-subject variability. Electroenceph. clin. Neurophysiol., 1987, 67: 402–411. Yang, J.F. and Winter, D.A. Electromyographic amplitude normalisation methods: improving their sensitivity as diagnostic tool in gait analysis. Arch. Phys. Med. Rehabil., 1984, 65: 517–521.
