- Email: [email protected]
A comparison of computer-based methods for the determination of onset of muscle contraction using electromyography
ELSEVIER
Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
A comparison of computer-based methods for the determination of onset of muscle contraction using electromyography Paul W. Hodges*, Bang H. Bui Department of Physiotherapy, The University of Queensland, Brisbane, Queensland 4072, Australia
Accepted for publication: 21 June 1996
Abstract Little consensus exists in the literature regarding methods for determination of the onset of electromyographic (EMG) activity. The aim of this study was to compare the relative accuracy of a range of computer-based techniques with respect to EMG onset determined visually by an experienced examiner. Twenty-seven methods were compared which varied in terms of EMG processing (low pass filtering at 10, 50 and 500 Hz), threshold value (1, 2 and 3 SD beyond mean of baseline activity) and the number of samples for which the mean must exceed the defined threshold (20, 50 and 100 ms). Three hundred randomly selected trials of a postural task were evaluated using each technique. The visual determination of EMG onset was found to be highly repeatable between days. Linear regression equations were calculated for the values selected by each computer method which indicated that the onset values selected by the majority of the parameter combinations deviated significantly from the visually derived onset values. Several methods accurately selected the time of onset of EMG activity and are recommended for future use. Copyright © 1996 Elsevier Science Ireland Ltd. Keywords: Electromyogram (EMG); Onset; Timing; Processing; Analysis
1. Introduction The temporal characteristics of electromyographic (EMG) recordings are parameters of neuromuscular function commonly used in the evaluation of posture and movement (e.g. Mann and Hagy, 1980; Cordo and Nashner, 1982; Carey et al., 1983; Lee et al., 1987; Latash et al., 1995). The onset of E M G is one of the most c o m m o n of these parameters evaluated; however, no standard method of determination of this parameter is used in the literature. In limb movement studies, where the reaction time of the postural muscles may be up to several hundred milliseconds, differences in the time from stimulus to E M G onset with neurological diseases, ageing and postural set m a y be as low as 20 ms (Traub et al., 1980; Cordo and Nashner, 1982; Woollacott and Manchester, 1993). This indicates that in order to allow comparisons between muscles, experimental conditions and subjects or subject groups, accuracy of onset determination is crucial. * Corresponding author. Tel.: +61 7 33652275; fax: +61 7 33652775; e-mail: [email protected]
The majority of studies evaluating the temporal parameters of E M G do not report the methods used for the identification of E M G onset (e.g. Belen'kii et al., 1967; Bouisset and Zattara, 1981; Oddsson and Thorstensson, 1987). In studies where the E M G onset determination method is described this is usually performed by visual evaluation of the E M G trace (e.g. Woollacott et al., 1988; Latash et al., 1995), generally without reporting the criteria on which this visually determined decision is made. Several studies report criteria such as the earliest detectable rise in E M G activity above the steady state (e.g. Allum and Pfaltz, 1985; Crenna et al., 1987; Woollacott et al., 1988; Inglis et al., 1994) or the point where the signal first deviates more than 1 or 1.5 SD from the level recorded during the steady state (Nashner et al., 1983; Nashner and Forssberg, 1986) without mention of how this is determined. Although the nature of visual onset determination is subjective, several authors report the inaccuracy to be as low as -+5-10 ms, without discussion of how this was assessed (Hallett et al., 1975; Horak et al., 1984). This inaccuracy is up to half the experimental difference
/96/$15.00 Copyright © 1996 Elsevier Science Ireland Ltd. All rights reserved PII S0921-884X(96)95190-5
951 90
512
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Table 1 Summary of onset determination criteria Author
Burst
Baseline
Signal processing
Method
Neafsey et al. (1978)
Two consecutive 50 ms bins >2 SD from baseline activity RMS>RMS threshold for >2 windows of 50 ms (threshold criteria not stated) First deviation >1.5 SD from baseline >3 SD beyond baseline for 25 ms >2 SD beyond baseline for 40 ms (both mean and median of sample) >2.5 SD beyond baseline
Initial 500 ms of 2s before stimulus -
Computer-based
100 ms prior to stimulus 50 ms prior to stimulus 500 ms quiet standing prior to stimulus 100 ms quiet stance
Averaged in 50 ms epochs 1000 Hz low pass filter 40 Hz low pass filter 50 Hz low pass filter 10-100 Hz band pass filter ? Hz low pass filter
3 x baseline for 8 of 16 ms Segment of n samples>previous segment at 5% significance level and following sample>previous at 5% significance level and current sample and following sample>30% maximal signal (n = 20% of movement duration) 15% maximal contraction for cycle of gait 5% of peak magnitude of burst
100 ms quiet standing ?
500 Hz low pass filter
Not stated Not stated
?
?
Not stated
500 ms prior to burst
>1 SD beyond baseline for 80 ms (not below for >16 ms) >1 SD beyond baseline, visually verified on computer screen and modified if inaccurate >95% confidence interval for >10 ms
240 ms prior to signal to move First 15 ms of sampling window
Low pass filter, subtraction Computer-based of baseline mean 6 Hz low pass filter Computer-based
Greenisen et al. (1979) Nashner et al. (1983) DiFabio (1987) Lee et al. (1987) Cfianand and Macpberson (1991) Studenski et al. (1991) Happee (1992)
Bullock-Saxton et al. (1993) Bullock-Saxton (1994) Steele (1994) Karst and Willet (1995) Thompson and McKinley (1995)
reported p r e v i o u s l y (Traub et al., 1980; C o r d o and Nashner, 1982; W o o l l a c o t t and Manchester, 1993). In several studies, traces w e r e assessed by two e x a m i n e r s and e x c l u d e d if the identified onsets differed by greater than 5 ms, p r o v i d i n g s o m e assurance o f reliability ( B r o w n and Frank, 1987; C r e n n a et al., 1987). A l t h o u g h , intra-tester reliability coefficients o f up to 0 . 7 8 - 0 . 8 2 h a v e b e e n identified w h e n c o m p a r i n g the visually derived E M G onsets identified by 3 e x p e r i e n c e d examiners, the p e r c e n t a g e o f trials in w h i c h the s a m e m i l l i s e c o n d was c h o s e n as the E M G onset was b e t w e e n 3 0 - 3 6 % w h e n c o m p a r i n g e x a m iners and b e t w e e n 4 7 - 5 6 % for each e x a m i n e r b e t w e e n days (DiFabio, 1987). Clearly, m u c h variation in E M G onset determination occurs w h e n this is done visually and is d e p e n d e n t on the e x p e r i e n c e and skill o f the e x a m iner. In an attempt to increase the objectivity (DiFabio, 1987) o f the evaluation o f E M G onset and to reduce o b s e r v e r bias (Studenski et al., 1991) an increasing n u m b e r o f studies rely on c o m p u t e r analysis (Table 1). Currently, there is little a g r e e m e n t regarding the m o s t appropriate method. A l t h o u g h each o f the c o m p u t e r algorithms use differing criteria to d e t e r m i n e the E M G onset, no studies h a v e evaluated the relative accuracy o f each or identified if specific trace characteristics require different onset determination methods.
-
Computer-based Not stated Computer-based Computer-based Not stated
Ensemble average of four trials
Computer-based
10-500 Hz band pass filter
Not stated
The aim o f the current study was to c o m p a r e a variety of criteria used for the c o m p u t e r d e t e r m i n a t i o n o f E M G onset against the E M G onsets d e t e r m i n e d visually by an experie n c e d e x a m i n e r to identify the parameters which p r o v i d e d the m o s t accurate identification o f E M G onset. A l t h o u g h it is a c k n o w l e d g e d that visual d e t e r m i n a t i o n o f E M G onset is not optimal, it currently p r o v i d e s the only standard against w h i c h to c o m p a r e the c o m p u t e r methods, hence the purpose of the study. Furthermore, the study aimed to identify the appropriate c o m b i n a t i o n o f parameters to m o s t accurately d e t e r m i n e E M G onset with a variety o f trace types. Identification of a c o m p u t e r algorithm that accurately identifies the onset o f E M G will provide a m e t h o d to increase the objectivity o f E M G onset determination that is i n d e p e n d e n t o f the e x p e r i e n c e or bias of the examiner.
2. Methods and materials 2.1. S u b j e c t s
The study i n v o l v e d a r a n d o m sample o f recordings of E M G activity c o l l e c t e d for another study (Hodges and Richardson, in preparation), i n v o l v i n g 4 subjects ( 1 8 - 2 4 years old) o f a v e r a g e height ( 1 . 5 9 - 1 . 8 5 m) and weight ( 5 5 - 7 5 kg). The study was a p p r o v e d by the M e d i c a l
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Research Ethics Committee of The University of Queensland. 2.2. Electromyographic recordings
The electromyographic activity of a selection of trunk and limb muscles was recorded using a combination of surface and fine-wire electrodes during a series of rapid upper limb movements. Fine wire electrodes were fabricated from Teflon-coated stainless steel wire (75/~m) with 1 m m of wire exposed. The bent tips of the wires were staggered at 1 and 3 m m to help maintain separation between the exposed ends. Electrodes were inserted into transversus abdominis, obliquus abdominis internus, obliquus abdominis externus and posterior gluteus medius. Ag/AgC1 surface electrodes were placed in parallel with the muscle fibres with an interelectrode separation of 12 m m over rectus abdominis, the anterior, middle and posterior portions of the deltoid muscle and the lumbar multifidus. Prior to surface electrode attachment the skin was prepared using a standard method to reduce the skin impedance to below 5 kf~. In a standing position subjects performed 10 repetitions each of rapid shoulder flexion, abduction and extension in response to a visual stimulus. The movement stimulus was preceded by a visual warning by a random period of between 0.5 and 4 s. E M G signals were bandpass filtered at 1 0 - 1 0 0 0 Hz (Basmajian and Deluca, 1985; Perry, 1992) and sampled at a rate of 2000 Hz with 12 bit analog to digital conversion using an A M L A B workstation (Associative Measurements, Australia). All data was saved on computer for further analysis. 2.3. Evaluation o f movement onset and duration
The duration of the movement of the limb was evaluated to provide a relative indication of the magnitude of the error associated with each computer-based criterion compared to the duration of the movement task. The duration of the movement was measured using a Position Velocity Transducer (UniMeasure, USA) which measures the velocity and distance of distraction of a cable from the apparatus during limb movement. The cable was attached to the wrist at a distance of 0.5 m from the approximate centre of rotation of the shoulder. The data was sampled at 100 Hz and stored on computer for later analysis. For the purpose of the current study the time from the stimulus to the onset of movement and the duration of the movement were evaluated. 2.4. Electromyographic onset determination
Three hundred samples were randomly selected from the available data. Data represented a wide spread of degree of background activity, magnitude of burst and
513
duration of burst and were separated into two categories of 150 traces each on the basis of background activity. All data with background activity greater than 20% of the mean of the rectified burst were categorised into the high background activity group. All traces were evaluated both visually and by a series of computer-based onset determination methods. If no onset could be determined by any particular method it was recorded as a missing value for the statistical analysis. Data in which the E M G onset was obscured by either movement artifact or a heart beat were eliminated so that the ability of the methods to identify true E M G onset could be evaluated. All E M G traces were full wave rectified prior to analysis. 2.4.1. Visual onset determination An experienced examiner visually determined the onset of E M G activity on the basis of the earliest rise in E M G activity beyond the steady state (Horak and Nashner, 1986; Brown and Frank, 1987). The raw E M G data was not further low pass filtered after the initial 10-1000 Hz band pass filtering which has been recommended previously to facilitate the visual determination (Dick et al., 1986; Latash et al., 1995). This was done to avoid biasing the agreement of the visually determined onsets with computer-based methods using similar data smoothing. All trials were displayed on a computer screen using interactive graphical software enabling the examiner to determine the onset to the nearest millisecond. Each muscle was displayed separately on the screen with no reference to the start of the movement to remove potential bias of the examiner. The examiner calculated the onset times for all traces on 2 separate days, 3 days apart, to evaluate the repeatability of the visual recordings. The mean of the visually determined onset times between days was used for evaluation of the computer-based methods. 2.4.2. Computer-based onset determination Twenty-seven different combinations of parameters were evaluated in the study. The basic algorithm involved identification of the point where the mean of a specified number of samples exceeded the baseline activity level (averaged for the 50 ms prior to the warning stimulus) by a specified number of standard deviations. The point at which the magnitude of the E M G trace exceeded this preset threshold was determined using a sliding window, the width of which was varied. The time of the initial sample of the window was recorded if the mean of the samples in the window exceeded the threshold. If the mean of the samples in the window did not exceed the criteria the window was advanced one sample at a time until an onset was found. If no onset was found within 2 s (4000 samples) a missing value was recorded for that trace. The parameters evaluated corresponded to those previously described in the literature (see Table 1). Firstly, the degree of smoothing of the E M G was varied using a sixth order elliptical low pass software filter at each of 3
514
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Table 2 Results of the linear regression analysis WW
10 10 10 10 10 10 10 10 10 25 25 25 25 25 25 25 25 25 50 50 50 50 50 50 50 50 50
SD
1 1 1 2 2 2 3 3 3 1 1 1 2 2 2 3 3 3 1 1 1 2 2 2 3 3 3
LPF
10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500
Total
Low
High
y
MV
y
MV
y
MV
-144.87'* -127.18"* 7.81 -126.36"* -89.01"* 60.64** -117.41"* -61.80"* 81.39"* -131.89"* -50.28** 119.15"* -113.31"* -20.6* 120.54"* -97.38** 5.51 146.73"* -87.21"* 1.39 142.88"* -70.32** 26.43* 149.17"* -56.44** 42.81"* 153.49"
0 0 0 0 0 13 0 0 26 0 0 59 0 0 91 0 1 121 0 1 178 0 10 220 0 19 240
-144.69"* -123.69"* -14.86 -127.59"* -88.47** 27.16" -119.61"* ~4.77"* 53.67** -128.16"* -51.25" 92.35** -111.72"* -21.55" 85.03** -97.02** -2.98 119.59"* -72.86** -4.42 152.16"* -57.93* 14.08 153.36"* -43.72* 29.16" 128.59"
0 0 0 0 0 2 0 0 3 0 0 8 0 0 19 0 0 33 0 0 55 0 2 84 0 4 98
-146.44"* -132.52"* 27.83* -126.28"* -90.89 90.91"* -116.52"* -58.95** 106.08"* -136.90"* -51.82"* 150.43"* -115.61"* -21.90 172.51"* -98.95** 11.71 188.73"* -101.05"* 5.89 123.76"* -82.37** 37.47* 196.78" -68.91"* 54.58* 475.87*
0 0 1 0 0 11 0 0 23 0 0 51 0 0 72 0 1 88 0 1 123 0 8 136 0 15 142
w w , width of window of samples required to exceed threshold (ms); SD, standard deviation; LPF, low pass filter (Hz); Y, y-intercept; MV, missing value. *P < 0.05, **P < 0.001. different values (10, 50, 500 Hz). A l t h o u g h low pass filtering at 10 Hz w o u l d r e m o v e m u c h o f the information of the trace after the initial high pass filtration at this frequency, the principle o f filtering at or b e l o w the initial high pass filtration has b e e n e x p o u n d e d by several authors (Dick et al., 1986; Steele, 1994) and was i n c l u d e d to e v a l u a t e the effect o f e x c e s s i v e filtering. In contrast, the upper freq u e n c y limit o f 500 H z is higher than the m a j o r i t y o f the signal d e t e c t e d by surface electrodes, due to low pass filtration by the tissues (Perry, 1992), resulting in m i n i m a l further s m o o t h i n g o f the data. This c o n d i t i o n was included to e v a l u a t e the effect o f inadequate filtering. The other parameters e v a l u a t e d w e r e the n u m b e r o f samples assessed in the sliding w i n d o w (20, 50, 100 samples or 10, 25, 50 ms) and the m a g n i t u d e of the deviation f r o m the baseline r e q u i r e d to indicate the threshold (1, 2, 3 SD). All 27 possible c o m b i n a t i o n s w e r e evaluated. P r o c e s s i n g was done using M A T L A B ' s signal processing toolkit (The MathWorks, USA). 2.5. Statistical analysis To e v a l u a t e the repeatabilty of the visually d e t e r m i n e d onset times b e t w e e n days the coefficient o f variation (CV)
and root m e a n square error ( R M S E ) were calculated. For the c o m p a r i s o n o f the visual and c o m p u t e r d e t e r m i n e d onset times the correlation b e t w e e n the two m e t h o d s was calculated. P e a r s o n ' s r determines the closeness o f fit o f the t w o variables to a linear relationship, i.e. linear regression line ( W e a v e r , 1983), and is unable to discriminate b e t w e e n relative amplitudes o f the two variables. Therefore, linear regression equations w e r e calculated using the m e a n o f the visually d e t e r m i n e d onset times b e t w e e n days as the i n d e p e n d e n t variable to identify the degree to w h i c h the c o m p u t e r - d e t e r m i n e d values d e v i a t e d f r o m those d e t e r m i n e d visually. All analyses w e r e undertaken on the data by group (low and high b a c k g r o u n d activity) and with the data f r o m both groups c o m b i n e d .
3. Results 3.1. Visual determination repeatability The repeatability of the visually d e t e r m i n e d E M G onset times b e t w e e n days was found to be high with a C V o f 0.42 and a R M S E o f 24.79. The repeatabilty o f the visually d e t e r m i n e d data for the low back ground activity and high b a c k g r o u n d activity were e v a l u a t e d separately, hav-
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
ing CV of 0.52 and 0.26, respectively, and RMSE of 31.69 and 15.00, respectively. 3.2. Visual determination versus computer determination The correlation between the computer-derived EMG onsets and the visually determined EMG onsets was extremely high for all methods (r = 0.999, P < 0.0001). Since the slope of the regression line (r) approached 1.0 for all of the methods evaluated, the y-intercept value can be considered to accurately depict the degree to which the computer-derived value (Oc) varies from the value determined visually (Or), using the equation of the regression line: O¢ = 0.999Ov + y-intercept. Therefore, the value of the y-intercept indicates the mean value that must be added to the visually determined onset time to approximate the value that would be determined by computer. The y-intercepts calculated for all 27 of the evaluated parameter combinations is presented in Table 2. The y-intercept for the majority of methods was significantly different from zero. Three combinations of parameters resulted in regression lines with a y-intercept that did not significantly deviate from the visually derived data (50 ms/1 SD/50 Hz, 25 ms/3 SD/50 Hz and 10 ms/1 SD/ 500 Hz). The y-intercept for the rest of the parameter combinations ranged from 20.6 to 153.49 ms (ignoring the sign of the value). The number of traces in which no onset could be determined by each method was also calculated (Table 2). The number of trials rejected was zero for the majority of trials but was as great as 240 when the criteria were most conservative (50 ms/3 SD/500 Hz). 3.3. Background activity level comparison When the computer and visually derived EMG onsets were compared for the low and high background activity groups separately there was some overlap between the methods with y-intercepts not significantly different from zero (Table 2). The 25 ms/3 SD/50 Hz and 50 ms/1 SD/50 Hz parameter combinations satisfied this criteria for both groups. For the low background activity group the 50 ms/2 SD/50 Hz and 10 ms/1 SD/500 Hz parameter combinations also satisfied this criteria as did 25 ms/2 SD/50 Hz for the high background activity group. The number of trials rejected with no discernible onset was greatest for the high background activity group. This occurred most frequently for the 50 ms/3 SD/500 Hz parameter combination, rejecting 142 of the 150 trials in this group. 3.4. Movement reaction time and duration The mean _+ SD of the reaction time (stimulus to movement initiation) for the movement of the limb was
515
285 + 33 ms and the mean movement duration (movement initiation to completion) was 194 _+ 36 ms. Therefore, the mean duration of the recording from stimulus to the completion of movement was 479 ms. The errors in determination of the EMG onset ranged from 7 to 167% of the reaction time and 4 to 99% of the total movement time. 4. Discussion The results indicate that the EMG onsets determined by the computer using the majority of the parameter combinations assessed vary significantly from the EMG onsets selected by visual determination. Although two parameter combinations accurately approximated the onset times regardless of the level of background, it was more common for the measurement to be less accurate when the background activity was high. Care should be used when selecting the appropriate analysis technique since errors of up to 167% of the reaction time may result from inappropriate methods, which is far in excess of the differences in EMG onset previously reported between postural conditions, age groups and pathologies (Traub et al., 1980; Cordo and Nashner, 1982; Woollacott and Manchester, 1993). The high correlation between the visually derived values and those identified using each of the computer methods indicates that the associated error is systematic, i.e. all computer-derived EMG onsets using a particular method vary from the visually derived values by the same amount. Accordingly, comparisons between muscles with similar background EMG activity is reasonable with any method since the error is identical. In contrast, when relating the onset of EMG to a time or movement event or when muscles with differing levels of background EMG activity are to be compared the systematic error is critical and the combination of parameters should be considered. 4.1. Influence of parameters on onset determination Low pass filtration is performed to remove the higher frequency components of the signal producing a smoother trace to assist the identification of the EMG onset (Soderberg and Cook, 1984) (Fig. 1). However, excessive smoothing of the data results in a loss of information and inaccurate identification of EMG onset (Halbertsma and DeBoer, 1981; Soderberg and Cook, 1984; Gabel and Brandt, 1994). Correspondingly, the errors associated with low pass filtration at a frequency identical to the lower limit of the initial bandpass filtering (10 Hz) in the current study resulted in errors of 56.44-144.87 ms prior to the visually determined value. Care must be taken when considering the results of studies where excessive smoothing of the data is undertaken (e.g. Dick et al., 1986; Steele, 1994). In contrast, insufficient data smoothing results in delayed identification of EMG onset due to the high frequency changes in amplitude, resulting in a reduced mean
516
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
o
0
0'.5
1.0 Time (s)
1.5
2.0
Fig. 1. Representative EMG traces from a single trial indicating effect of low pass filtering. (A) Full wave rectified raw EMG; (B) 500 Hz low pass filter; (C) 50 Hz low pass filter; (D) 10 Hz low pass filter. The onset determined using the 25 m and 1 SD criteria is marked for each filter type as well as the visually determined onset for the full wave rectified raw EMG trace. Note the inaccuracy in onset determination resulting from the different levels of smoothing of the data. for the specified number of samples. This effect was consistently noted for the 500 Hz low pass filtration condition, producing errors of 60.64-153.49 ms after the visually derived E M G onset. The number of trials in which no E M G onset was identified was also large for this condition. This was with the exception of the least conservative condition where the threshold amplitude was low and the window period was short (10 ms/1 SD/500 Hz), producing values not significantly different from those identified visually. On the basis of these findings, the results of previous studies incorporating low pass filtration at 500 Hz and a threshold of 2 SD (95% confidence interval) are likely to be associated with error (Happee, 1992; Thompson and McKinley, 1995). The low pass cut-off frequency that most accurately and consistently determined the E M G onset in the current study was 50 Hz. Several previous studies have incorporated filtration similar to this value (Nashner et al., 1983; DiFabio, 1987). Little consensus exists for the threshold criteria for computer-based E M G onset determination. Statistically based E M G onset determination methods report changes of 1 - 3 SD which influences the probability of making a Type I or II error. Type I errors occur when the muscle is identified as active when it is not, which arises if the threshold is low, such as the 1 SD condition which is the point beyond which 15.87% of the instantaneous background activity occurs. In contrast, Type II errors indicate a failure to identify the E M G onset when it occurs. The frequency of a type II error is increased in the conservative 3 SD condition. Although the results of the current study indicate delayed E M G onset determination with the 3 SD condition relative to the other threshold criteria, individual parameter combinations involving each threshold level were found to be non-significantly different from the
visually derived values and previously reported methods cannot be disputed on the basis of the threshold criteria. The number of milliseconds for which the mean must exceed the threshold also influenced the accuracy of onset determination. When the sample is short the chance of identifying an erratic burst of background activity as the onset is increased. In contrast, when the sample is long the chance of ignoring a short E M G burst is increased. Furthermore, when the E M G onset has a steep slope and large amplitude the threshold mean may be exceeded when the initial data sample, which is recorded as the onset, is several milliseconds from the onset. In the present study the short sample width (10 ms) consistently identified the E M G onset prior to the visually derived value by a mean of 46.31 ms, the long sample width (50 ms) delayed the identification of onset of E M G by a mean of 33.58 ms and the 25 ms sample width closely approximated the visually derived value with a mean error of 3.16 ms. Previous studies using unnecessarily long or short sample widths should be evaluated with caution (Table 1). The criteria that most accurately identified the onset of E M G (25 ms/3 SD/50 Hz, 50 ms/1 SD/50 Hz, 10 ms/1 SD/ 500 Hz) resulted from the most appropriate combinations of the positive and negative influences of each of the parameters. The accuracy of the 25 ms/3 SD/50 Hz criteria has previously been evaluated against visual identification of E M G onset, indicating that the millisecond identified by computer corresponded to that identified visually in only 3 0 - 3 6 % of the trials (DiFabio, 1987). In the current study the computer-derived values deviated by 5.51 ms from those identified visually, providing an explanation for the poor agreement described by DiFabio (1987). With the exception of the parameter combinations used by DiFabio (1987) and Lee et al. (1987), the majority of the methods described in the literature are likely to have resulted in inaccuracies of onset determination (Table 1). 4.2. Influence o f the signal-to-noise ratio
The benefit of a threshold based on the statistical deviation from the baseline is that it is normalised to the erratic changes in baseline activity, which is rarely silent in postural tasks. Consequently, the onset value identified by computer for traces with high background activity was consistently delayed relative to the visually determined value (Fig. 2). Although it was anticipated that different parameter combinations would be ideal for each background activity level group, there was considerable overlap between the most accurate parameter combinations for each group (Table 2). Another method used to reduce the influence of background activity is the subtraction of the mean of the baseline activity level from the trace (Bullock-Saxton, 1994). This method would fail to remove erratic bursts of baseline activity of larger amplitude than the mean. Although several researchers have recommended instructing the subject
517
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Several authors have suggested techniques to average (Bogey et al., 1992; Gitter, 1994) or summate (Ellaway, 1978) EMG traces to determine a representative timing profile that is less influenced by random variation. The accuracy of these techniques has not been evaluated against visual determination of onset time.
1
,
L
4.4. Considerations for computer determination of EMG onset
12
0~s
1~o
Time (s)
1~5
2.o
Fig. 2. Variation in the EMG onset time determined by computer analysis using the most (10 Hz/1 SD/10 ms) (1) and least accurate (50 Hz/l SD/50 ms) (2) methods for traces with (A) high background activity, and (B) low background activity. Note the large variation for the high background activity condition.
to consciously relax the muscles of interest prior to data collection to limit the influence of this problem (Lee, 1980; Badke and DiFabio, 1985), potential changes in behaviour caused by this intervention need to be considered. The most accurate comparison between muscles is achieved when the characteristics of the EMG traces are similar. When the trace characteristics vary widely, particularly in terms of the signal-to-noise ratio, then the use of separate onset determination criteria should be considered. 4.3. Other criteria involved in EMG onset determination Several studies have included further criteria to minimise the chance of type I errors such as dictating that the median value (Lee et al., 1987) or a minimum number of samples within the window are required to exceed the threshold (Studenski et al., 1991; Steele, 1994). The addition of these criteria would assist in the rejection of erratic short duration bursts of background activity. Assignment of the middle sample of the window rather than the initial sample as the onset may reduce the influence of slope of EMG increase. In contrast to the statistically based threshold criteria, several authors have developed methods based on the time taken to reach a predetermined EMG amplitude (Greenisen et al., 1979; Bullock-Saxton et al., 1993; BullockSaxton, 1994). Identification of the EMG onset as the point where the EMG amplitude reaches 5% of the peak value following prior subtraction of the background activity level (Bullock-Saxton et al., 1993; Bullock-Saxton, 1994) is potentially inaccurate due to the sensitivity of this technique to the magnitude of the peak and the rate of increase in amplitude which are likely to vary between muscles. The onset of EMG would be delayed when the EMG trace is of large amplitude and when the rate of increase is gradual. Other methods in which a minimum magnitude criteria is used would also be associated with delayed EMG onset identification (Happee, 1992).
Interference of EMG traces by movement artifact or electrocardiograms may obscure the onset of EMG activity (Bruce et al., 1977) (Fig. 3). For this reason, any completely automated method for identification of EMG onset needs to recognise and reject traces in which the EMG onset is confused by this interference. Currently, no method exists with the complex pattern recognition skill required to perform this task. Until this is achieved it is necessary to visually check each trace against the computer-derived value to ensure that the onset is meaningful. In the current study, traces in which the onset was obscured were excluded so that the accuracy of each method could be determined for the evaluation of true EMG onset. Furthermore, visual verification of the trace can ensure that the onset of EMG has not be overlooked if the burst duration is short or of low magnitude (Neafsey et al., 1978; Lee et al., 1987). Finally, regardless of the criteria used the onset of EMG determined by computer is likely to be influenced by the rate of increase in EMG amplitude (Walter, 1984), such as slow movements, which would result in a delay in onset identification (Horak et al., 1984; DiFabio, 1987). It is recommended that special care be taken with the visual verification of computer-determined onset times with this type of data and lower threshold values may be beneficial. Further evaluation of threshold levels for the determination of onset with this type of data would be beneficial. 4.5. Conclusions and recommendations In summary, several combinations of parameters have been identified that allow for accurate determination of EMG onset using computer-based algorithmic calculations. The advantage of these methods is an increased objectivity of analysis, reduced time required to perform the analysis and a reduced requirement for experimenter
0
0.5
1.0
1,5
Time (s)
2.0
2.5
Fig. 3. Presentation of heart ECG (electrocardiogram) interfering with the EMG onset determination.
518
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
e x p e r i e n c e , as the o n l y r e q u i r e m e n t is to e n s u r e that m o v e m e n t artifact or o t h e r i n t e r f e r e n c e is n o t w r o n g l y determ i n e d as the E M G onset. It is r e c o m m e n d e d that the c h a r a c t e r i s t i c s o f the E M G trace s h o u l d b e c o n s i d e r e d w h e n s e l e c t i n g the a p p r o p r i a t e p a r a m e t e r c o m b i n a t i o n for analysis.
Acknowledgements W e t h a n k Mr. I. H o r t o n for statistical a d v i c e and c o m m e n t s o n the m a n u s c r i p t , Dr. B. R i c h a r d s o n and Dr. D. C o o p e r for s u p e r v i s i o n o f the f i n e - w i r e e l e c t r o d e insertions a n d Dr. C. R i c h a r d s o n for useful c o m m e n t s o n the manuscript.
References Allure, J.H.J. and Pfaltz, C.R. Visual and vestibular contributions to pitch sway stabilisation in the ankle muscles of normals and patients with bilateral peripheral vestibular deficits. Exp. Brain Res., 1985, 58: 82-94. Badke, M.B. and DiFabio, R.P. Effects of postural bias during support surface displacements and rapid arm movements. Phys. Ther., 1985, 65: 1490-1495. Basmajian, J.V. and Deluca, C. Muscles Alive: Their Function Revealed by Electromyography. Williams and Wilkins, Baltimore, 1985. Belen'kii, V., Gurfinkel, V.S. and Paltsev, Y. Elements of control of voluntary movements. Biofizika, 1967, 12: 135-141. Bogey, R.A., Barnes, L.A. and Perry, J. Computer algorithms to characterize individual subject EMG profiles during gait. Arch. Phys. Med. Rehabil., 1992, 73: 835-841. Bouisset, S. and Zattara, M.A. Sequence of postural adjustments precedes voluntary movement. Neurosei. Lett., 1981, 22: 263-270. Brown, J. and Frank, J.S. Influence of event anticipation on postural actions accompanying voluntary movement. Exp. Brain Res., 1987, 67: 645-650. Bruce, E.N., Goldman, M.D. and Mead, J. A digital computer technique for analyzing respiratory muscle EMGs. J. Appl. Physiol., 1977, 43: 551-556. Bullock-Saxton, J.E. Local sensation changes and altered hip muscle function following severe ankle sprain. Phys. Ther., 1994, 74: 17-31. Bullock-Saxton, J.E., Janda, V. and Bullock, M.I. Reflex activation of gluteal muscles in walking with balance shoes: an approach to restoration of function for chronic low back pain patients. Spine, 1993, 18: 704-708. Carey, J.R., Allison, J.D. and Mundale, M.O. Electromyographic study of muscular overflow during precision hand grip. Phys. Ther., 1983, 63:505-511. Chanaud, C.M. and Macpherson, J.M. Functionally complex muscles of the cat hindlimb, III: differential activation within biceps femoris during postural perturbation. Exp. Brain Res., 1991, 85: 271-280. Cordo, P.J. and Nashner, L.M. Properties of postural adjustments associated with rapid ann movements. J. Neurophysiol., 1982, 47: 287308. Crenna, P., Frigo, C., Massion, J. and Pedotti, A. Forward and backward axial synergies in man. Exp. Brain Res., 1987, 65: 538-548. Dick, J.P.R., Rothwell, J.C., Berardelli, A., Thompson, P.D., Gioux, M., Benecke, R., Day, B.L. and Marsden, C.D. Associated postural adjustments in Parkinson's disease. J. Neurol. Neurosurg. Psychiatry, 1986, 49: 1378-1385. DiFabio, R.P. Reliability of computerised surface electromyography for determining the onset of muscle activity. Phys. Ther., 1987, 67: 4348. Ellaway, P.H. Cumulative sum technique and its applications to the
analysis of prestimulus time histograms. Electroenceph. clin. Neurophysiol., 1978, 45: 302-304. Gabel, R.H. and Brandt, R.A. The effect of signal conditioning on statistical analyses of gait EMG. Electroenceph. clin. Neurophysiol., 1994, 93: 188-201. Gitter, A. Advances in electrodiagnostic instruments: principles and applications of digital signal processing. Phys. Med. Rehabil. Clin. North Am., 1994, 5: 397-419. Greenisen, M.C., Vroomen, L.J. and Vroomen, L.C. Computerised interpretation of initiation and relaxation of EMG recorded dynamic muscle contraction. Proceedings of the 4th Congress of the International Society of Electrophysiological Kinesiology, Boston, USA, 1979, pp. 172-173. Halbertsma, J.M. and DeBoer, R.R. On the processing of electromyograms for computer analysis. J. Biomech., 1981, 14: 431-435. Hallett, M., Shahani, B.T. and Young, R.R. EMG analysis of stereotyped voluntary movements in man. J. Neurol. Neurosurg. Psychiatry, 1975, 38: 1154-1162. Happee, R. Goal-directed ann movements, I: analysis of EMG records in shoulder and elbow muscles. J. Electromyogr. Kinesiol., 1992, 2: 165-178. Horak, F. and Nashner, L.M. Central programming of postural movements: adaptation to altered support-surface configurations. J. Neurophysiol., 1986, 55: 1369-1381. Horak, F.B., Esselman, P., Anderson, M.E. and Lynch, M.K. The effects of movement velocity, mass displaced, and task certainty on associated postural adjustments made by normal and hemiplegic individuals. J. Neurol. Neurosurg. Psychiatry, 1984, 47: 1020-1028. Inglis, J.T., Horak, F.B., Shupert, C.L. and Jones-Rycewicz, C. The importance of somatosensory information in triggering and scaling automatic postural responses in humans. Exp. Brain Res., 1994, 101: 159-164. Karst, G.M. and Willet, G.M. Onset timing of electromyographic activity in the vastus medialis oblique and vastus lateralis muscles in subjects with and without patellofemoral pain syndrome. Phys. Ther., 1995, 75: 813-823. Latash, M.L., Aruin, A.S., Neyman, I. and Nichols, J.J. Anticipatory postural adjustments during self-inflicted and predictable perturbations in Parkinson's disease. J. Neurol. Neurosurg. Psychiatry, 1995, 58: 326-334. Lee, W.A. Anticipatory control of postural and task muscles during rapid arm flexion. J. Mot. Behav., 1980, 12: 185-196. Lee, W.A., Buchanan, T.S. and Rogers, M.W. Effects of ann acceleration and behavioural conditions on the organisation of postural adjustments during arm flexion. Exp. Brain Res., 1987, 66: 257-270. Mann, R.A. and Hagy, J. Biomechanics of walking, running, and sprinting. Am. J. Sports Med., 1980, 8: 345-350. Nashner, L.M. and Forssberg, H. Phase-dependent organisation of postural adjustments associated with arm movements while walking. J. Neurophysiol., 1986, 55: 1382-1394. Nashner, L.M., Shumway-Cook, A. and Marin, O. Stance posture control in select groups of children with cerebral palsy: deficits in sensory organisation and muscular coordination. Exp. Brain Res., 1983, 49: 393-409. Neafsey, E.J., Hull, C.D. and Buchwald, N.A. Preparation for movement in the cat, II: unit activity in the basal ganglia and thalamus. Electroenceph, clin. Neurophysiol., 1978, 44: 714-723. Oddsson, L. and Thorstensson, A. Fast voluntary trunk flexion movements in standing: motor patterns. Acta Physiol. Scand., 1987, 129: 93-106. Perry, J. Gait Analysis: Normal and Pathological Function. Slack Incorporated, Thorofare, NJ, 1992. Soderberg, G.L. and Cook, T.M. Electromyography in biomechanics. Phys. Ther., 1984, 64: 1813-1820. Steele, C. Anticipatory postural reactions in clumsy children: changes after physiotherapy treatment. Master of Physiotherapy Thesis, University of Queensland, 1994.
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519 Studenski, S., Duncan, P.W. and Chandler, J. Postural responses and effector factors in persons with unexplained falls: results and methodological issues. J. Am. Geriatr. Soc., 1991, 39: 229-234. Thompson, H.W. and McKinley, P.A. Landing from a jump: the role of vision when landing from known and unknown heights. NeuroReport, 1995, 6: 581-584. Traub, M.M., Rothwell, J.C. and Marsden, C.D. Anticipatory postural reflexes in Parkinson's disease and other akinetic-rigid syndromes and in cerebellar ataxia. Brain, 1980, 103: 393-412. Walter, C.B. Temporal quantification of electromyography with
519
reference to motor control research. Hum. Move. Sci., 1984, 3: 155-162. Weaver, J.H. Applications of Discrete and Continuous Fourier Analysis. Wiley, New York, 1983. Woollacott, M.H. and Manchester, D.L. Anticipatory postural adjustments in older adults: are changes in response characteristics due to changes in strategy. J. Gerontol. Med. Sci., 1993, 48: 64-70. Woollacott, M.H., von Hosten, C. and R6sblad, B. Relation between muscle response onset and body segmental movements during postural perturbations in humans. Exp. Brain Res., 1988, 72: 593-604.
Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
A comparison of computer-based methods for the determination of onset of muscle contraction using electromyography Paul W. Hodges*, Bang H. Bui Department of Physiotherapy, The University of Queensland, Brisbane, Queensland 4072, Australia
Accepted for publication: 21 June 1996
Abstract Little consensus exists in the literature regarding methods for determination of the onset of electromyographic (EMG) activity. The aim of this study was to compare the relative accuracy of a range of computer-based techniques with respect to EMG onset determined visually by an experienced examiner. Twenty-seven methods were compared which varied in terms of EMG processing (low pass filtering at 10, 50 and 500 Hz), threshold value (1, 2 and 3 SD beyond mean of baseline activity) and the number of samples for which the mean must exceed the defined threshold (20, 50 and 100 ms). Three hundred randomly selected trials of a postural task were evaluated using each technique. The visual determination of EMG onset was found to be highly repeatable between days. Linear regression equations were calculated for the values selected by each computer method which indicated that the onset values selected by the majority of the parameter combinations deviated significantly from the visually derived onset values. Several methods accurately selected the time of onset of EMG activity and are recommended for future use. Copyright © 1996 Elsevier Science Ireland Ltd. Keywords: Electromyogram (EMG); Onset; Timing; Processing; Analysis
1. Introduction The temporal characteristics of electromyographic (EMG) recordings are parameters of neuromuscular function commonly used in the evaluation of posture and movement (e.g. Mann and Hagy, 1980; Cordo and Nashner, 1982; Carey et al., 1983; Lee et al., 1987; Latash et al., 1995). The onset of E M G is one of the most c o m m o n of these parameters evaluated; however, no standard method of determination of this parameter is used in the literature. In limb movement studies, where the reaction time of the postural muscles may be up to several hundred milliseconds, differences in the time from stimulus to E M G onset with neurological diseases, ageing and postural set m a y be as low as 20 ms (Traub et al., 1980; Cordo and Nashner, 1982; Woollacott and Manchester, 1993). This indicates that in order to allow comparisons between muscles, experimental conditions and subjects or subject groups, accuracy of onset determination is crucial. * Corresponding author. Tel.: +61 7 33652275; fax: +61 7 33652775; e-mail: [email protected]
The majority of studies evaluating the temporal parameters of E M G do not report the methods used for the identification of E M G onset (e.g. Belen'kii et al., 1967; Bouisset and Zattara, 1981; Oddsson and Thorstensson, 1987). In studies where the E M G onset determination method is described this is usually performed by visual evaluation of the E M G trace (e.g. Woollacott et al., 1988; Latash et al., 1995), generally without reporting the criteria on which this visually determined decision is made. Several studies report criteria such as the earliest detectable rise in E M G activity above the steady state (e.g. Allum and Pfaltz, 1985; Crenna et al., 1987; Woollacott et al., 1988; Inglis et al., 1994) or the point where the signal first deviates more than 1 or 1.5 SD from the level recorded during the steady state (Nashner et al., 1983; Nashner and Forssberg, 1986) without mention of how this is determined. Although the nature of visual onset determination is subjective, several authors report the inaccuracy to be as low as -+5-10 ms, without discussion of how this was assessed (Hallett et al., 1975; Horak et al., 1984). This inaccuracy is up to half the experimental difference
/96/$15.00 Copyright © 1996 Elsevier Science Ireland Ltd. All rights reserved PII S0921-884X(96)95190-5
951 90
512
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Table 1 Summary of onset determination criteria Author
Burst
Baseline
Signal processing
Method
Neafsey et al. (1978)
Two consecutive 50 ms bins >2 SD from baseline activity RMS>RMS threshold for >2 windows of 50 ms (threshold criteria not stated) First deviation >1.5 SD from baseline >3 SD beyond baseline for 25 ms >2 SD beyond baseline for 40 ms (both mean and median of sample) >2.5 SD beyond baseline
Initial 500 ms of 2s before stimulus -
Computer-based
100 ms prior to stimulus 50 ms prior to stimulus 500 ms quiet standing prior to stimulus 100 ms quiet stance
Averaged in 50 ms epochs 1000 Hz low pass filter 40 Hz low pass filter 50 Hz low pass filter 10-100 Hz band pass filter ? Hz low pass filter
3 x baseline for 8 of 16 ms Segment of n samples>previous segment at 5% significance level and following sample>previous at 5% significance level and current sample and following sample>30% maximal signal (n = 20% of movement duration) 15% maximal contraction for cycle of gait 5% of peak magnitude of burst
100 ms quiet standing ?
500 Hz low pass filter
Not stated Not stated
?
?
Not stated
500 ms prior to burst
>1 SD beyond baseline for 80 ms (not below for >16 ms) >1 SD beyond baseline, visually verified on computer screen and modified if inaccurate >95% confidence interval for >10 ms
240 ms prior to signal to move First 15 ms of sampling window
Low pass filter, subtraction Computer-based of baseline mean 6 Hz low pass filter Computer-based
Greenisen et al. (1979) Nashner et al. (1983) DiFabio (1987) Lee et al. (1987) Cfianand and Macpberson (1991) Studenski et al. (1991) Happee (1992)
Bullock-Saxton et al. (1993) Bullock-Saxton (1994) Steele (1994) Karst and Willet (1995) Thompson and McKinley (1995)
reported p r e v i o u s l y (Traub et al., 1980; C o r d o and Nashner, 1982; W o o l l a c o t t and Manchester, 1993). In several studies, traces w e r e assessed by two e x a m i n e r s and e x c l u d e d if the identified onsets differed by greater than 5 ms, p r o v i d i n g s o m e assurance o f reliability ( B r o w n and Frank, 1987; C r e n n a et al., 1987). A l t h o u g h , intra-tester reliability coefficients o f up to 0 . 7 8 - 0 . 8 2 h a v e b e e n identified w h e n c o m p a r i n g the visually derived E M G onsets identified by 3 e x p e r i e n c e d examiners, the p e r c e n t a g e o f trials in w h i c h the s a m e m i l l i s e c o n d was c h o s e n as the E M G onset was b e t w e e n 3 0 - 3 6 % w h e n c o m p a r i n g e x a m iners and b e t w e e n 4 7 - 5 6 % for each e x a m i n e r b e t w e e n days (DiFabio, 1987). Clearly, m u c h variation in E M G onset determination occurs w h e n this is done visually and is d e p e n d e n t on the e x p e r i e n c e and skill o f the e x a m iner. In an attempt to increase the objectivity (DiFabio, 1987) o f the evaluation o f E M G onset and to reduce o b s e r v e r bias (Studenski et al., 1991) an increasing n u m b e r o f studies rely on c o m p u t e r analysis (Table 1). Currently, there is little a g r e e m e n t regarding the m o s t appropriate method. A l t h o u g h each o f the c o m p u t e r algorithms use differing criteria to d e t e r m i n e the E M G onset, no studies h a v e evaluated the relative accuracy o f each or identified if specific trace characteristics require different onset determination methods.
-
Computer-based Not stated Computer-based Computer-based Not stated
Ensemble average of four trials
Computer-based
10-500 Hz band pass filter
Not stated
The aim o f the current study was to c o m p a r e a variety of criteria used for the c o m p u t e r d e t e r m i n a t i o n o f E M G onset against the E M G onsets d e t e r m i n e d visually by an experie n c e d e x a m i n e r to identify the parameters which p r o v i d e d the m o s t accurate identification o f E M G onset. A l t h o u g h it is a c k n o w l e d g e d that visual d e t e r m i n a t i o n o f E M G onset is not optimal, it currently p r o v i d e s the only standard against w h i c h to c o m p a r e the c o m p u t e r methods, hence the purpose of the study. Furthermore, the study aimed to identify the appropriate c o m b i n a t i o n o f parameters to m o s t accurately d e t e r m i n e E M G onset with a variety o f trace types. Identification of a c o m p u t e r algorithm that accurately identifies the onset o f E M G will provide a m e t h o d to increase the objectivity o f E M G onset determination that is i n d e p e n d e n t o f the e x p e r i e n c e or bias of the examiner.
2. Methods and materials 2.1. S u b j e c t s
The study i n v o l v e d a r a n d o m sample o f recordings of E M G activity c o l l e c t e d for another study (Hodges and Richardson, in preparation), i n v o l v i n g 4 subjects ( 1 8 - 2 4 years old) o f a v e r a g e height ( 1 . 5 9 - 1 . 8 5 m) and weight ( 5 5 - 7 5 kg). The study was a p p r o v e d by the M e d i c a l
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Research Ethics Committee of The University of Queensland. 2.2. Electromyographic recordings
The electromyographic activity of a selection of trunk and limb muscles was recorded using a combination of surface and fine-wire electrodes during a series of rapid upper limb movements. Fine wire electrodes were fabricated from Teflon-coated stainless steel wire (75/~m) with 1 m m of wire exposed. The bent tips of the wires were staggered at 1 and 3 m m to help maintain separation between the exposed ends. Electrodes were inserted into transversus abdominis, obliquus abdominis internus, obliquus abdominis externus and posterior gluteus medius. Ag/AgC1 surface electrodes were placed in parallel with the muscle fibres with an interelectrode separation of 12 m m over rectus abdominis, the anterior, middle and posterior portions of the deltoid muscle and the lumbar multifidus. Prior to surface electrode attachment the skin was prepared using a standard method to reduce the skin impedance to below 5 kf~. In a standing position subjects performed 10 repetitions each of rapid shoulder flexion, abduction and extension in response to a visual stimulus. The movement stimulus was preceded by a visual warning by a random period of between 0.5 and 4 s. E M G signals were bandpass filtered at 1 0 - 1 0 0 0 Hz (Basmajian and Deluca, 1985; Perry, 1992) and sampled at a rate of 2000 Hz with 12 bit analog to digital conversion using an A M L A B workstation (Associative Measurements, Australia). All data was saved on computer for further analysis. 2.3. Evaluation o f movement onset and duration
The duration of the movement of the limb was evaluated to provide a relative indication of the magnitude of the error associated with each computer-based criterion compared to the duration of the movement task. The duration of the movement was measured using a Position Velocity Transducer (UniMeasure, USA) which measures the velocity and distance of distraction of a cable from the apparatus during limb movement. The cable was attached to the wrist at a distance of 0.5 m from the approximate centre of rotation of the shoulder. The data was sampled at 100 Hz and stored on computer for later analysis. For the purpose of the current study the time from the stimulus to the onset of movement and the duration of the movement were evaluated. 2.4. Electromyographic onset determination
Three hundred samples were randomly selected from the available data. Data represented a wide spread of degree of background activity, magnitude of burst and
513
duration of burst and were separated into two categories of 150 traces each on the basis of background activity. All data with background activity greater than 20% of the mean of the rectified burst were categorised into the high background activity group. All traces were evaluated both visually and by a series of computer-based onset determination methods. If no onset could be determined by any particular method it was recorded as a missing value for the statistical analysis. Data in which the E M G onset was obscured by either movement artifact or a heart beat were eliminated so that the ability of the methods to identify true E M G onset could be evaluated. All E M G traces were full wave rectified prior to analysis. 2.4.1. Visual onset determination An experienced examiner visually determined the onset of E M G activity on the basis of the earliest rise in E M G activity beyond the steady state (Horak and Nashner, 1986; Brown and Frank, 1987). The raw E M G data was not further low pass filtered after the initial 10-1000 Hz band pass filtering which has been recommended previously to facilitate the visual determination (Dick et al., 1986; Latash et al., 1995). This was done to avoid biasing the agreement of the visually determined onsets with computer-based methods using similar data smoothing. All trials were displayed on a computer screen using interactive graphical software enabling the examiner to determine the onset to the nearest millisecond. Each muscle was displayed separately on the screen with no reference to the start of the movement to remove potential bias of the examiner. The examiner calculated the onset times for all traces on 2 separate days, 3 days apart, to evaluate the repeatability of the visual recordings. The mean of the visually determined onset times between days was used for evaluation of the computer-based methods. 2.4.2. Computer-based onset determination Twenty-seven different combinations of parameters were evaluated in the study. The basic algorithm involved identification of the point where the mean of a specified number of samples exceeded the baseline activity level (averaged for the 50 ms prior to the warning stimulus) by a specified number of standard deviations. The point at which the magnitude of the E M G trace exceeded this preset threshold was determined using a sliding window, the width of which was varied. The time of the initial sample of the window was recorded if the mean of the samples in the window exceeded the threshold. If the mean of the samples in the window did not exceed the criteria the window was advanced one sample at a time until an onset was found. If no onset was found within 2 s (4000 samples) a missing value was recorded for that trace. The parameters evaluated corresponded to those previously described in the literature (see Table 1). Firstly, the degree of smoothing of the E M G was varied using a sixth order elliptical low pass software filter at each of 3
514
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Table 2 Results of the linear regression analysis WW
10 10 10 10 10 10 10 10 10 25 25 25 25 25 25 25 25 25 50 50 50 50 50 50 50 50 50
SD
1 1 1 2 2 2 3 3 3 1 1 1 2 2 2 3 3 3 1 1 1 2 2 2 3 3 3
LPF
10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500 10 50 500
Total
Low
High
y
MV
y
MV
y
MV
-144.87'* -127.18"* 7.81 -126.36"* -89.01"* 60.64** -117.41"* -61.80"* 81.39"* -131.89"* -50.28** 119.15"* -113.31"* -20.6* 120.54"* -97.38** 5.51 146.73"* -87.21"* 1.39 142.88"* -70.32** 26.43* 149.17"* -56.44** 42.81"* 153.49"
0 0 0 0 0 13 0 0 26 0 0 59 0 0 91 0 1 121 0 1 178 0 10 220 0 19 240
-144.69"* -123.69"* -14.86 -127.59"* -88.47** 27.16" -119.61"* ~4.77"* 53.67** -128.16"* -51.25" 92.35** -111.72"* -21.55" 85.03** -97.02** -2.98 119.59"* -72.86** -4.42 152.16"* -57.93* 14.08 153.36"* -43.72* 29.16" 128.59"
0 0 0 0 0 2 0 0 3 0 0 8 0 0 19 0 0 33 0 0 55 0 2 84 0 4 98
-146.44"* -132.52"* 27.83* -126.28"* -90.89 90.91"* -116.52"* -58.95** 106.08"* -136.90"* -51.82"* 150.43"* -115.61"* -21.90 172.51"* -98.95** 11.71 188.73"* -101.05"* 5.89 123.76"* -82.37** 37.47* 196.78" -68.91"* 54.58* 475.87*
0 0 1 0 0 11 0 0 23 0 0 51 0 0 72 0 1 88 0 1 123 0 8 136 0 15 142
w w , width of window of samples required to exceed threshold (ms); SD, standard deviation; LPF, low pass filter (Hz); Y, y-intercept; MV, missing value. *P < 0.05, **P < 0.001. different values (10, 50, 500 Hz). A l t h o u g h low pass filtering at 10 Hz w o u l d r e m o v e m u c h o f the information of the trace after the initial high pass filtration at this frequency, the principle o f filtering at or b e l o w the initial high pass filtration has b e e n e x p o u n d e d by several authors (Dick et al., 1986; Steele, 1994) and was i n c l u d e d to e v a l u a t e the effect o f e x c e s s i v e filtering. In contrast, the upper freq u e n c y limit o f 500 H z is higher than the m a j o r i t y o f the signal d e t e c t e d by surface electrodes, due to low pass filtration by the tissues (Perry, 1992), resulting in m i n i m a l further s m o o t h i n g o f the data. This c o n d i t i o n was included to e v a l u a t e the effect o f inadequate filtering. The other parameters e v a l u a t e d w e r e the n u m b e r o f samples assessed in the sliding w i n d o w (20, 50, 100 samples or 10, 25, 50 ms) and the m a g n i t u d e of the deviation f r o m the baseline r e q u i r e d to indicate the threshold (1, 2, 3 SD). All 27 possible c o m b i n a t i o n s w e r e evaluated. P r o c e s s i n g was done using M A T L A B ' s signal processing toolkit (The MathWorks, USA). 2.5. Statistical analysis To e v a l u a t e the repeatabilty of the visually d e t e r m i n e d onset times b e t w e e n days the coefficient o f variation (CV)
and root m e a n square error ( R M S E ) were calculated. For the c o m p a r i s o n o f the visual and c o m p u t e r d e t e r m i n e d onset times the correlation b e t w e e n the two m e t h o d s was calculated. P e a r s o n ' s r determines the closeness o f fit o f the t w o variables to a linear relationship, i.e. linear regression line ( W e a v e r , 1983), and is unable to discriminate b e t w e e n relative amplitudes o f the two variables. Therefore, linear regression equations w e r e calculated using the m e a n o f the visually d e t e r m i n e d onset times b e t w e e n days as the i n d e p e n d e n t variable to identify the degree to w h i c h the c o m p u t e r - d e t e r m i n e d values d e v i a t e d f r o m those d e t e r m i n e d visually. All analyses w e r e undertaken on the data by group (low and high b a c k g r o u n d activity) and with the data f r o m both groups c o m b i n e d .
3. Results 3.1. Visual determination repeatability The repeatability of the visually d e t e r m i n e d E M G onset times b e t w e e n days was found to be high with a C V o f 0.42 and a R M S E o f 24.79. The repeatabilty o f the visually d e t e r m i n e d data for the low back ground activity and high b a c k g r o u n d activity were e v a l u a t e d separately, hav-
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
ing CV of 0.52 and 0.26, respectively, and RMSE of 31.69 and 15.00, respectively. 3.2. Visual determination versus computer determination The correlation between the computer-derived EMG onsets and the visually determined EMG onsets was extremely high for all methods (r = 0.999, P < 0.0001). Since the slope of the regression line (r) approached 1.0 for all of the methods evaluated, the y-intercept value can be considered to accurately depict the degree to which the computer-derived value (Oc) varies from the value determined visually (Or), using the equation of the regression line: O¢ = 0.999Ov + y-intercept. Therefore, the value of the y-intercept indicates the mean value that must be added to the visually determined onset time to approximate the value that would be determined by computer. The y-intercepts calculated for all 27 of the evaluated parameter combinations is presented in Table 2. The y-intercept for the majority of methods was significantly different from zero. Three combinations of parameters resulted in regression lines with a y-intercept that did not significantly deviate from the visually derived data (50 ms/1 SD/50 Hz, 25 ms/3 SD/50 Hz and 10 ms/1 SD/ 500 Hz). The y-intercept for the rest of the parameter combinations ranged from 20.6 to 153.49 ms (ignoring the sign of the value). The number of traces in which no onset could be determined by each method was also calculated (Table 2). The number of trials rejected was zero for the majority of trials but was as great as 240 when the criteria were most conservative (50 ms/3 SD/500 Hz). 3.3. Background activity level comparison When the computer and visually derived EMG onsets were compared for the low and high background activity groups separately there was some overlap between the methods with y-intercepts not significantly different from zero (Table 2). The 25 ms/3 SD/50 Hz and 50 ms/1 SD/50 Hz parameter combinations satisfied this criteria for both groups. For the low background activity group the 50 ms/2 SD/50 Hz and 10 ms/1 SD/500 Hz parameter combinations also satisfied this criteria as did 25 ms/2 SD/50 Hz for the high background activity group. The number of trials rejected with no discernible onset was greatest for the high background activity group. This occurred most frequently for the 50 ms/3 SD/500 Hz parameter combination, rejecting 142 of the 150 trials in this group. 3.4. Movement reaction time and duration The mean _+ SD of the reaction time (stimulus to movement initiation) for the movement of the limb was
515
285 + 33 ms and the mean movement duration (movement initiation to completion) was 194 _+ 36 ms. Therefore, the mean duration of the recording from stimulus to the completion of movement was 479 ms. The errors in determination of the EMG onset ranged from 7 to 167% of the reaction time and 4 to 99% of the total movement time. 4. Discussion The results indicate that the EMG onsets determined by the computer using the majority of the parameter combinations assessed vary significantly from the EMG onsets selected by visual determination. Although two parameter combinations accurately approximated the onset times regardless of the level of background, it was more common for the measurement to be less accurate when the background activity was high. Care should be used when selecting the appropriate analysis technique since errors of up to 167% of the reaction time may result from inappropriate methods, which is far in excess of the differences in EMG onset previously reported between postural conditions, age groups and pathologies (Traub et al., 1980; Cordo and Nashner, 1982; Woollacott and Manchester, 1993). The high correlation between the visually derived values and those identified using each of the computer methods indicates that the associated error is systematic, i.e. all computer-derived EMG onsets using a particular method vary from the visually derived values by the same amount. Accordingly, comparisons between muscles with similar background EMG activity is reasonable with any method since the error is identical. In contrast, when relating the onset of EMG to a time or movement event or when muscles with differing levels of background EMG activity are to be compared the systematic error is critical and the combination of parameters should be considered. 4.1. Influence of parameters on onset determination Low pass filtration is performed to remove the higher frequency components of the signal producing a smoother trace to assist the identification of the EMG onset (Soderberg and Cook, 1984) (Fig. 1). However, excessive smoothing of the data results in a loss of information and inaccurate identification of EMG onset (Halbertsma and DeBoer, 1981; Soderberg and Cook, 1984; Gabel and Brandt, 1994). Correspondingly, the errors associated with low pass filtration at a frequency identical to the lower limit of the initial bandpass filtering (10 Hz) in the current study resulted in errors of 56.44-144.87 ms prior to the visually determined value. Care must be taken when considering the results of studies where excessive smoothing of the data is undertaken (e.g. Dick et al., 1986; Steele, 1994). In contrast, insufficient data smoothing results in delayed identification of EMG onset due to the high frequency changes in amplitude, resulting in a reduced mean
516
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
o
0
0'.5
1.0 Time (s)
1.5
2.0
Fig. 1. Representative EMG traces from a single trial indicating effect of low pass filtering. (A) Full wave rectified raw EMG; (B) 500 Hz low pass filter; (C) 50 Hz low pass filter; (D) 10 Hz low pass filter. The onset determined using the 25 m and 1 SD criteria is marked for each filter type as well as the visually determined onset for the full wave rectified raw EMG trace. Note the inaccuracy in onset determination resulting from the different levels of smoothing of the data. for the specified number of samples. This effect was consistently noted for the 500 Hz low pass filtration condition, producing errors of 60.64-153.49 ms after the visually derived E M G onset. The number of trials in which no E M G onset was identified was also large for this condition. This was with the exception of the least conservative condition where the threshold amplitude was low and the window period was short (10 ms/1 SD/500 Hz), producing values not significantly different from those identified visually. On the basis of these findings, the results of previous studies incorporating low pass filtration at 500 Hz and a threshold of 2 SD (95% confidence interval) are likely to be associated with error (Happee, 1992; Thompson and McKinley, 1995). The low pass cut-off frequency that most accurately and consistently determined the E M G onset in the current study was 50 Hz. Several previous studies have incorporated filtration similar to this value (Nashner et al., 1983; DiFabio, 1987). Little consensus exists for the threshold criteria for computer-based E M G onset determination. Statistically based E M G onset determination methods report changes of 1 - 3 SD which influences the probability of making a Type I or II error. Type I errors occur when the muscle is identified as active when it is not, which arises if the threshold is low, such as the 1 SD condition which is the point beyond which 15.87% of the instantaneous background activity occurs. In contrast, Type II errors indicate a failure to identify the E M G onset when it occurs. The frequency of a type II error is increased in the conservative 3 SD condition. Although the results of the current study indicate delayed E M G onset determination with the 3 SD condition relative to the other threshold criteria, individual parameter combinations involving each threshold level were found to be non-significantly different from the
visually derived values and previously reported methods cannot be disputed on the basis of the threshold criteria. The number of milliseconds for which the mean must exceed the threshold also influenced the accuracy of onset determination. When the sample is short the chance of identifying an erratic burst of background activity as the onset is increased. In contrast, when the sample is long the chance of ignoring a short E M G burst is increased. Furthermore, when the E M G onset has a steep slope and large amplitude the threshold mean may be exceeded when the initial data sample, which is recorded as the onset, is several milliseconds from the onset. In the present study the short sample width (10 ms) consistently identified the E M G onset prior to the visually derived value by a mean of 46.31 ms, the long sample width (50 ms) delayed the identification of onset of E M G by a mean of 33.58 ms and the 25 ms sample width closely approximated the visually derived value with a mean error of 3.16 ms. Previous studies using unnecessarily long or short sample widths should be evaluated with caution (Table 1). The criteria that most accurately identified the onset of E M G (25 ms/3 SD/50 Hz, 50 ms/1 SD/50 Hz, 10 ms/1 SD/ 500 Hz) resulted from the most appropriate combinations of the positive and negative influences of each of the parameters. The accuracy of the 25 ms/3 SD/50 Hz criteria has previously been evaluated against visual identification of E M G onset, indicating that the millisecond identified by computer corresponded to that identified visually in only 3 0 - 3 6 % of the trials (DiFabio, 1987). In the current study the computer-derived values deviated by 5.51 ms from those identified visually, providing an explanation for the poor agreement described by DiFabio (1987). With the exception of the parameter combinations used by DiFabio (1987) and Lee et al. (1987), the majority of the methods described in the literature are likely to have resulted in inaccuracies of onset determination (Table 1). 4.2. Influence o f the signal-to-noise ratio
The benefit of a threshold based on the statistical deviation from the baseline is that it is normalised to the erratic changes in baseline activity, which is rarely silent in postural tasks. Consequently, the onset value identified by computer for traces with high background activity was consistently delayed relative to the visually determined value (Fig. 2). Although it was anticipated that different parameter combinations would be ideal for each background activity level group, there was considerable overlap between the most accurate parameter combinations for each group (Table 2). Another method used to reduce the influence of background activity is the subtraction of the mean of the baseline activity level from the trace (Bullock-Saxton, 1994). This method would fail to remove erratic bursts of baseline activity of larger amplitude than the mean. Although several researchers have recommended instructing the subject
517
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
Several authors have suggested techniques to average (Bogey et al., 1992; Gitter, 1994) or summate (Ellaway, 1978) EMG traces to determine a representative timing profile that is less influenced by random variation. The accuracy of these techniques has not been evaluated against visual determination of onset time.
1
,
L
4.4. Considerations for computer determination of EMG onset
12
0~s
1~o
Time (s)
1~5
2.o
Fig. 2. Variation in the EMG onset time determined by computer analysis using the most (10 Hz/1 SD/10 ms) (1) and least accurate (50 Hz/l SD/50 ms) (2) methods for traces with (A) high background activity, and (B) low background activity. Note the large variation for the high background activity condition.
to consciously relax the muscles of interest prior to data collection to limit the influence of this problem (Lee, 1980; Badke and DiFabio, 1985), potential changes in behaviour caused by this intervention need to be considered. The most accurate comparison between muscles is achieved when the characteristics of the EMG traces are similar. When the trace characteristics vary widely, particularly in terms of the signal-to-noise ratio, then the use of separate onset determination criteria should be considered. 4.3. Other criteria involved in EMG onset determination Several studies have included further criteria to minimise the chance of type I errors such as dictating that the median value (Lee et al., 1987) or a minimum number of samples within the window are required to exceed the threshold (Studenski et al., 1991; Steele, 1994). The addition of these criteria would assist in the rejection of erratic short duration bursts of background activity. Assignment of the middle sample of the window rather than the initial sample as the onset may reduce the influence of slope of EMG increase. In contrast to the statistically based threshold criteria, several authors have developed methods based on the time taken to reach a predetermined EMG amplitude (Greenisen et al., 1979; Bullock-Saxton et al., 1993; BullockSaxton, 1994). Identification of the EMG onset as the point where the EMG amplitude reaches 5% of the peak value following prior subtraction of the background activity level (Bullock-Saxton et al., 1993; Bullock-Saxton, 1994) is potentially inaccurate due to the sensitivity of this technique to the magnitude of the peak and the rate of increase in amplitude which are likely to vary between muscles. The onset of EMG would be delayed when the EMG trace is of large amplitude and when the rate of increase is gradual. Other methods in which a minimum magnitude criteria is used would also be associated with delayed EMG onset identification (Happee, 1992).
Interference of EMG traces by movement artifact or electrocardiograms may obscure the onset of EMG activity (Bruce et al., 1977) (Fig. 3). For this reason, any completely automated method for identification of EMG onset needs to recognise and reject traces in which the EMG onset is confused by this interference. Currently, no method exists with the complex pattern recognition skill required to perform this task. Until this is achieved it is necessary to visually check each trace against the computer-derived value to ensure that the onset is meaningful. In the current study, traces in which the onset was obscured were excluded so that the accuracy of each method could be determined for the evaluation of true EMG onset. Furthermore, visual verification of the trace can ensure that the onset of EMG has not be overlooked if the burst duration is short or of low magnitude (Neafsey et al., 1978; Lee et al., 1987). Finally, regardless of the criteria used the onset of EMG determined by computer is likely to be influenced by the rate of increase in EMG amplitude (Walter, 1984), such as slow movements, which would result in a delay in onset identification (Horak et al., 1984; DiFabio, 1987). It is recommended that special care be taken with the visual verification of computer-determined onset times with this type of data and lower threshold values may be beneficial. Further evaluation of threshold levels for the determination of onset with this type of data would be beneficial. 4.5. Conclusions and recommendations In summary, several combinations of parameters have been identified that allow for accurate determination of EMG onset using computer-based algorithmic calculations. The advantage of these methods is an increased objectivity of analysis, reduced time required to perform the analysis and a reduced requirement for experimenter
0
0.5
1.0
1,5
Time (s)
2.0
2.5
Fig. 3. Presentation of heart ECG (electrocardiogram) interfering with the EMG onset determination.
518
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519
e x p e r i e n c e , as the o n l y r e q u i r e m e n t is to e n s u r e that m o v e m e n t artifact or o t h e r i n t e r f e r e n c e is n o t w r o n g l y determ i n e d as the E M G onset. It is r e c o m m e n d e d that the c h a r a c t e r i s t i c s o f the E M G trace s h o u l d b e c o n s i d e r e d w h e n s e l e c t i n g the a p p r o p r i a t e p a r a m e t e r c o m b i n a t i o n for analysis.
Acknowledgements W e t h a n k Mr. I. H o r t o n for statistical a d v i c e and c o m m e n t s o n the m a n u s c r i p t , Dr. B. R i c h a r d s o n and Dr. D. C o o p e r for s u p e r v i s i o n o f the f i n e - w i r e e l e c t r o d e insertions a n d Dr. C. R i c h a r d s o n for useful c o m m e n t s o n the manuscript.
References Allure, J.H.J. and Pfaltz, C.R. Visual and vestibular contributions to pitch sway stabilisation in the ankle muscles of normals and patients with bilateral peripheral vestibular deficits. Exp. Brain Res., 1985, 58: 82-94. Badke, M.B. and DiFabio, R.P. Effects of postural bias during support surface displacements and rapid arm movements. Phys. Ther., 1985, 65: 1490-1495. Basmajian, J.V. and Deluca, C. Muscles Alive: Their Function Revealed by Electromyography. Williams and Wilkins, Baltimore, 1985. Belen'kii, V., Gurfinkel, V.S. and Paltsev, Y. Elements of control of voluntary movements. Biofizika, 1967, 12: 135-141. Bogey, R.A., Barnes, L.A. and Perry, J. Computer algorithms to characterize individual subject EMG profiles during gait. Arch. Phys. Med. Rehabil., 1992, 73: 835-841. Bouisset, S. and Zattara, M.A. Sequence of postural adjustments precedes voluntary movement. Neurosei. Lett., 1981, 22: 263-270. Brown, J. and Frank, J.S. Influence of event anticipation on postural actions accompanying voluntary movement. Exp. Brain Res., 1987, 67: 645-650. Bruce, E.N., Goldman, M.D. and Mead, J. A digital computer technique for analyzing respiratory muscle EMGs. J. Appl. Physiol., 1977, 43: 551-556. Bullock-Saxton, J.E. Local sensation changes and altered hip muscle function following severe ankle sprain. Phys. Ther., 1994, 74: 17-31. Bullock-Saxton, J.E., Janda, V. and Bullock, M.I. Reflex activation of gluteal muscles in walking with balance shoes: an approach to restoration of function for chronic low back pain patients. Spine, 1993, 18: 704-708. Carey, J.R., Allison, J.D. and Mundale, M.O. Electromyographic study of muscular overflow during precision hand grip. Phys. Ther., 1983, 63:505-511. Chanaud, C.M. and Macpherson, J.M. Functionally complex muscles of the cat hindlimb, III: differential activation within biceps femoris during postural perturbation. Exp. Brain Res., 1991, 85: 271-280. Cordo, P.J. and Nashner, L.M. Properties of postural adjustments associated with rapid ann movements. J. Neurophysiol., 1982, 47: 287308. Crenna, P., Frigo, C., Massion, J. and Pedotti, A. Forward and backward axial synergies in man. Exp. Brain Res., 1987, 65: 538-548. Dick, J.P.R., Rothwell, J.C., Berardelli, A., Thompson, P.D., Gioux, M., Benecke, R., Day, B.L. and Marsden, C.D. Associated postural adjustments in Parkinson's disease. J. Neurol. Neurosurg. Psychiatry, 1986, 49: 1378-1385. DiFabio, R.P. Reliability of computerised surface electromyography for determining the onset of muscle activity. Phys. Ther., 1987, 67: 4348. Ellaway, P.H. Cumulative sum technique and its applications to the
analysis of prestimulus time histograms. Electroenceph. clin. Neurophysiol., 1978, 45: 302-304. Gabel, R.H. and Brandt, R.A. The effect of signal conditioning on statistical analyses of gait EMG. Electroenceph. clin. Neurophysiol., 1994, 93: 188-201. Gitter, A. Advances in electrodiagnostic instruments: principles and applications of digital signal processing. Phys. Med. Rehabil. Clin. North Am., 1994, 5: 397-419. Greenisen, M.C., Vroomen, L.J. and Vroomen, L.C. Computerised interpretation of initiation and relaxation of EMG recorded dynamic muscle contraction. Proceedings of the 4th Congress of the International Society of Electrophysiological Kinesiology, Boston, USA, 1979, pp. 172-173. Halbertsma, J.M. and DeBoer, R.R. On the processing of electromyograms for computer analysis. J. Biomech., 1981, 14: 431-435. Hallett, M., Shahani, B.T. and Young, R.R. EMG analysis of stereotyped voluntary movements in man. J. Neurol. Neurosurg. Psychiatry, 1975, 38: 1154-1162. Happee, R. Goal-directed ann movements, I: analysis of EMG records in shoulder and elbow muscles. J. Electromyogr. Kinesiol., 1992, 2: 165-178. Horak, F. and Nashner, L.M. Central programming of postural movements: adaptation to altered support-surface configurations. J. Neurophysiol., 1986, 55: 1369-1381. Horak, F.B., Esselman, P., Anderson, M.E. and Lynch, M.K. The effects of movement velocity, mass displaced, and task certainty on associated postural adjustments made by normal and hemiplegic individuals. J. Neurol. Neurosurg. Psychiatry, 1984, 47: 1020-1028. Inglis, J.T., Horak, F.B., Shupert, C.L. and Jones-Rycewicz, C. The importance of somatosensory information in triggering and scaling automatic postural responses in humans. Exp. Brain Res., 1994, 101: 159-164. Karst, G.M. and Willet, G.M. Onset timing of electromyographic activity in the vastus medialis oblique and vastus lateralis muscles in subjects with and without patellofemoral pain syndrome. Phys. Ther., 1995, 75: 813-823. Latash, M.L., Aruin, A.S., Neyman, I. and Nichols, J.J. Anticipatory postural adjustments during self-inflicted and predictable perturbations in Parkinson's disease. J. Neurol. Neurosurg. Psychiatry, 1995, 58: 326-334. Lee, W.A. Anticipatory control of postural and task muscles during rapid arm flexion. J. Mot. Behav., 1980, 12: 185-196. Lee, W.A., Buchanan, T.S. and Rogers, M.W. Effects of ann acceleration and behavioural conditions on the organisation of postural adjustments during arm flexion. Exp. Brain Res., 1987, 66: 257-270. Mann, R.A. and Hagy, J. Biomechanics of walking, running, and sprinting. Am. J. Sports Med., 1980, 8: 345-350. Nashner, L.M. and Forssberg, H. Phase-dependent organisation of postural adjustments associated with arm movements while walking. J. Neurophysiol., 1986, 55: 1382-1394. Nashner, L.M., Shumway-Cook, A. and Marin, O. Stance posture control in select groups of children with cerebral palsy: deficits in sensory organisation and muscular coordination. Exp. Brain Res., 1983, 49: 393-409. Neafsey, E.J., Hull, C.D. and Buchwald, N.A. Preparation for movement in the cat, II: unit activity in the basal ganglia and thalamus. Electroenceph, clin. Neurophysiol., 1978, 44: 714-723. Oddsson, L. and Thorstensson, A. Fast voluntary trunk flexion movements in standing: motor patterns. Acta Physiol. Scand., 1987, 129: 93-106. Perry, J. Gait Analysis: Normal and Pathological Function. Slack Incorporated, Thorofare, NJ, 1992. Soderberg, G.L. and Cook, T.M. Electromyography in biomechanics. Phys. Ther., 1984, 64: 1813-1820. Steele, C. Anticipatory postural reactions in clumsy children: changes after physiotherapy treatment. Master of Physiotherapy Thesis, University of Queensland, 1994.
P.W. Hodges, B.H. Bui / Electroencephalography and clinical Neurophysiology 101 (1996) 511-519 Studenski, S., Duncan, P.W. and Chandler, J. Postural responses and effector factors in persons with unexplained falls: results and methodological issues. J. Am. Geriatr. Soc., 1991, 39: 229-234. Thompson, H.W. and McKinley, P.A. Landing from a jump: the role of vision when landing from known and unknown heights. NeuroReport, 1995, 6: 581-584. Traub, M.M., Rothwell, J.C. and Marsden, C.D. Anticipatory postural reflexes in Parkinson's disease and other akinetic-rigid syndromes and in cerebellar ataxia. Brain, 1980, 103: 393-412. Walter, C.B. Temporal quantification of electromyography with
519
reference to motor control research. Hum. Move. Sci., 1984, 3: 155-162. Weaver, J.H. Applications of Discrete and Continuous Fourier Analysis. Wiley, New York, 1983. Woollacott, M.H. and Manchester, D.L. Anticipatory postural adjustments in older adults: are changes in response characteristics due to changes in strategy. J. Gerontol. Med. Sci., 1993, 48: 64-70. Woollacott, M.H., von Hosten, C. and R6sblad, B. Relation between muscle response onset and body segmental movements during postural perturbations in humans. Exp. Brain Res., 1988, 72: 593-604.