审批

2009.02.13
Journal of Neuroscience Methods 90 (1999) 47 – 55
bingrenyg

A comparison of cross-correlation and surface EMG techniques

used to quantify motor unit synchronization in humans
John G. Semmler, Michael A. Nordstrom *
Department of Physiology, Uni6ersity of Adelaide, Adelaide, SA 5005, Australia

Received 10 March 1999; accepted 8 May 1999

Abstract

Two methods used to estimate the strength of motor unit (MU) synchronization in a muscle are the direct cross-correlation of
MU discharge times, and averaging of the surface electromyogram (SEMG) with respect to discharge of a reference MU.
Although indirect, the latter approach has the advantage that a global estimate of MU synchrony can be obtained quickly and
easily. The two methods are generally regarded as providing equivalent information on the extent of MU synchronization in a
muscle, but this proposition has not previously been tested quantitatively. In the present study, we used both the SEMG technique
(189 MUs) and cross-correlation of MU discharge (498 MU pairs) to estimate MU synchrony in 28 first dorsal interosseus (FDI)
muscles from 16 subjects. Despite considerable overlap in the identity of MUs used to quantify synchrony with each method,
linear regression revealed no significant correlation between the estimates of MU synchronization in FDI muscles obtained with
the two techniques (r 2 =0.04, n=28). This discrepancy was not due to insufficient sampling of the MU population with the
cross-correlation method, although we found evidence for a non-uniform tendency for synchronous discharge in two of 13 motor
units providing sufficient data for the analysis. The most likely explanation for the discrepancy between the estimates of MU
synchrony is that methodological problems with the SEMG technique limit its accuracy. These problems are difficult to avoid
under normal experimental conditions, and we conclude that the SEMG method is not reliable for quantitative comparisons of
MU synchrony between muscles and subjects. © 1999 Elsevier Science B.V. All rights reserved.

Keywords: Cross-correlation; Motor unit; Surface electromyogram; Corticospinal tract; First dorsal interosseous; Synchrony

1. Introduction histogram constructed of the peri-event discharge times

During a voluntary isometric contraction there is an
increased tendency for motor units (MUs) to discharge
within a few milliseconds of each other more often than
is expected due to chance. This MU synchronization is
of interest because its features can be used to infer
details of the operation of last-order neurons with
widely divergent projections to motoneurons (Sears and
Stagg, 1976; Datta and Stephens, 1990). The corti-
cospinal projection appears to be of major importance
for MU synchronization in distal hand muscles (Datta
et al., 1991; Farmer et al., 1993). MU synchrony can be
quantified directly by cross-correlation of individual
discharge times of MU pairs, in which the discharge
times of one MU are used as a reference, and a
* Corresponding author. Tel.: +61-8-8303-4567; fax: + 61-8-8303-
3356.
of the other MU (Moore et al., 1966). Although this
direct method accurately quantifies synchronization of
MU pairs, in practice it poses considerable technical
difficulties. Short-term synchrony is fairly weak, and
trials of several minutes duration are needed to provide
sufficient counts for a reliable estimate of synchrony
from the cross-correlogram. It is difficult to identify the
same MUs in the intramuscular recordings from several
electrodes for long periods, as their action potential
waveforms may change. A further complication is that
two intramuscular electrodes are needed, and each must
sample from a separate population of MUs to avoid
spurious effects in the cross-correlogram due to non-
random discrimination errors arising from action po-
tential superimpositions involving an MU detected in
both channels. The large territory occupied by muscle
fibres of a single MU make this difficult to avoid in
small muscles. Finally, it is necessary to sample from a

0165-0270/99/$ - see front matter © 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 5 - 0 2 7 0 ( 9 9 ) 0 0 0 6 9 - 2
48 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
number of MU pairs to obtain a reliable estimate of the
overall extent of MU synchrony in a muscle, because
the strength of MU synchrony varies over a wide range
for different MU pairs in the same muscle (Bremner et
al., 1991; Nordstrom et al., 1992).
The technical challenges of the cross-correlation tech-
nique have prompted the use of alternative methods
which are quicker and easier to use. Milner-Brown et
al. (1973) introduced a technique for estimating the
global MU synchronization in a muscle which consists
of comparing the simultaneously recorded unrectified
and full-wave rectified surface EMG (SEMG) which
have been averaged with respect to the discharge of a
reference MU. Comparison of the areas of the rectified
and unrectified SEMG averages (above baseline levels)
provides an index of synchronization of reference MU
discharges with those of other active MUs. The SEMG
synchronization index derived using this method has
the advantage that all active MUs make some contribu-
tion to the SEMG signal, and so an analysis of a
relatively short period of discharge of the reference MU
provides a global estimate of its synchronization with
other MUs in the muscle. Synchronization indices can
be quickly and easily obtained for a number of refer-
ence MUs in a single session. Using this method, for
example, Milner-Brown et al. (1975) have reported that
strength-training enhances MU synchrony in the first
dorsal interosseous (FDI) muscle, a change which has
been attributed to strengthening of transcortical reflex
pathways.
It is generally accepted that the SEMG and MU
cross-correlogram techniques provide equivalent infor-
mation on the extent of MU synchronization in a
muscle. Surprisingly, however, this proposition has not
been quantitatively tested using experimental data. The
SEMG technique has recently been shown to be subject
to technical problems which raise questions about its
reliability for estimating MU synchronization (Yue et
al., 1995). In the present study we have obtained spike-
train data from an extensive sample of MUs in 28 FDI
muscles, and used these data to estimate MU syn-
chrony in these muscles with both the SEMG technique
(189 MUs) and cross-correlation of MU discharge (498
MU pairs). The results show that the two methods do
not produce equivalent estimates of the overall extent
of MU synchronization in a muscle.
2. Methods
All subjects volunteered to participate in the study
and gave informed consent to the procedures, which
were approved by the Human Research Ethics Com-
mittee at the University of Adelaide. MU activity was
recorded from 28 FDI muscles in five musicians, five
strength-trained and six untrained subjects (ages 18–47
years). Features of MU discharge, MU synchrony and
force tremor in these subjects have been reported previ-
ously (Semmler and Nordstrom, 1995, 1998). The ex-
tensive MU sample obtained in FDI muscles in these
subjects made the present investigation possible. As the
extent of FDI MU synchrony differs in the three
groups (Semmler and Nordstrom, 1998) these subjects
exhibit a large range of MU synchrony in FDI, which
improves the likelihood of demonstrating a significant
correlation between estimates of MU synchrony ob-
tained with the two methods.
The experimental arrangement and protocol for
recording the surface EMG and MU discharge proper-
ties have been described previously (Semmler and
Nordstrom, 1995, 1998). Briefly, the subjects attended
the laboratory on two or more separate occasions,
where SEMG and MU activity were recorded from one
hand on each occasion. The subjects’ right or left arm
and hand were secured in a manipulandum, and the
force of abduction was measured by a load cell which
was aligned with the distal interphalangeal joint of the
index finger. The SEMG of FDI was recorded with
bipolar Ag/AgCl electrodes placed 2–3 cm apart. MU
activity was recorded simultaneously with two separate
fine-wire electrodes (1–2 cm interelectrode distance)
which were inserted percutaneously into the FDI with a
25-gauge disposable needle. Myoelectric signals were
amplified (1000× ), filtered (bandwidth 2 Hz–10 kHz)
and recorded on FM tape (Vetter model 400D, Rebers-
burg, PA, USA, 22 kHz/ch) for off-line analysis. The
DC force signal was filtered (0–50 Hz) and digitised (1
kHz) on-line on a Macintosh computer.
To begin the experiment the subjects performed a
steady, low-force, isometric abduction of the index
finger. A single MU was chosen by the experimenters
for the subject to control at a comfortable discharge
rate (feedback unit). The subjects were provided with
audio and visual feedback of MU discharge on an
oscilloscope screen. The subject’s task was to control
the mean firing rate of the feedback unit at a constant
target level (usually 8, 10 or 12 Hz) for 1–5 min. The
activity of additional MUs was monitored during the
trial to confirm that discriminable MU potentials were
present in each channel for off-line cross-correlation.
The procedure was repeated following repositioning of
both electrodes, in order to sample from as many
different MUs as possible in the session.
All analyses were performed off-line from the taped
records. Single MUs from each intramuscular electrode
were discriminated using a computer-based template-
matching algorithm (SPS 8701, Signal Processing Sys-
tem, Malvern, S.A., Australia). Action potentials
belonging to a particular MU were identified on the
basis of waveform shape, and great care was taken to
confirm the identity of units discriminated during dif-
ferent trials (see Nordstrom et al., 1992). ISIs of iden-
 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55 49

tified MUs were measured (9250 ms resolution) using was above the baseline activity (black and hatched area,
an in-built function of the SPS 8701 and stored on Fig. 1B and C), and above the voltage levels which
computer. ISI records were scrutinised for every trial occurred in the unrectified EMG trace (hatched area
and each discriminated MU to assess discrimination only, Fig. 1B and C). The strength of MU synchroniza-
accuracy. ISI histograms were constructed from the tion was determined by computing the ratio of the
discharge times of each MU. ISI files with \ 5% dis- synchronous area (black area, Fig. 1B and C) to the
crimination error (usually missed spikes due to super- area of the unrectified EMG within the peak region
impositions) were excluded from all analyses. (dotted and hatched area, Fig. 1B and C).
For 189 MUs, TTL pulses corresponding to the time MU synchronization was also assessed using cross-
of discharge of a single MU discriminated with the SPS correlation of MU discharge times (Nordstrom et al.,
8701 were sent to a second computer with the corre-
1992). MUs detected with separate electrodes in the
sponding surface EMG (2 kHz sampling rate). The
single MU discharge times were used as the reference
(t =0 ms) for spike-triggered averaging (STA) of the
SEMG. STA of the digitally full-wave rectified and
unrectified EMG was performed with a custom-de-
signed computer program. Each STA had a duration of
9 85 ms around the time of discharge of the reference
MU, and was based on 150 reference MU discharges.
A comparison of the area of the unrectified and
rectified EMG averages was used to provide an esti-
mate of the strength of MU synchronization within the
MU population. This method has previously been de-
scribed by Milner-Brown and co-workers (Milner-
Brown et al., 1973, 1975) and is shown for two MUs in
Fig. 1B and C. Briefly, STA of the unrectified and
full-wave rectified SEMG was performed for each refer-
ence MU. The unrectified SEMG average essentially
represents the contribution of only the reference MU to
the SEMG, because the contribution of the positive and
negative waveform components from other MUs not
synchronized to the reference MU average to zero. The
rectified SEMG average provides a measure of the total
electrical activity in the muscle around the time of
reference MU discharge, which includes the waveform Fig. 1. Estimation of the strength of MU synchronization from the
of the reference unit after rectification, an average cross-correlogram and the corresponding SEMG average. (A) Esti-
EMG level (background) due to independent firing of mation of MU synchrony using the cross-correlogram of MU dis-
charge. The lower trace is the cross-correlogram of the discharge
other MUs, an artifact associated with signal rectifica-
times of two concurrently active MUs in FDI. The mean bin count of
tion (due to a partial summation of the rectified wave- off-peak bins was 9.7 (horizontal line). This value served to distin-
form and the background activity; Milner-Brown et al. guish between the counts expected due to chance (vertically hatched
(1973); see also Yue et al., 1995) and the contribution area) from those counts in excess of chance (black area) in the peak
of other MUs that are synchronized to the reference region. The position and duration of the synchronous peak was
judged visually from the cumulative sum (CUSUM, upper trace). In
MU. For each reference MU, the unrectified (with
this example, peak duration (vertical dashed lines) was 19 ms centred
negative phases inverted) and rectified EMG averages at t = −2 ms. The synchronization index CIS for this MU pair was
were superimposed and the boundaries for the area of 1.21 extra counts s − 1. (B, C) Estimation of MU synchrony using the
synchronous activity were identified (dashed vertical SEMG spike-triggered averaging method (n = 150) for the two MUs
lines, Fig. 1B and C). This consisted of detecting the in (A). The area of the unrectified EMG average is represented by the
first location either side of the central peak, starting at dotted and hatched area. The area of the rectified EMG average is
represented by the black and hatched area. The duration of the peak
the point of MU discharge (arrow, Fig. 1B and C), is designated by the dashed vertical lines. The extent of MU synchro-
where the rectified EMG level was equal to the mean nization is calculated from the ratio of the synchronized area (black
EMG level due to the asynchronous firing of other area) to the total area of the unrectified EMG average within the
MUs (i.e. background EMG level; dashed horizontal peak region (dotted and hatched area). The arrow indicates the time
lines, Fig. 1B and C). From within the peak region, the of discharge of the reference MU. (B) The SEMG synchronization
index for this MU is 0.31 with a peak duration of 19 ms. (C) Data
area due to the synchronous firing of other MUs was from the second concurrently active MU in (A). This MU was used
established (black area in Fig. 1B and C). This area as the reference MU in the cross-correlogram (A). SEMG synchro-
represented the area in the rectified EMG trace which nization index for this MU was 1.03 with a peak duration of 24 ms.
50 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
same trial were paired for cross-correlation (498 MU
pairs). All cross-correlation histograms had 1-ms bin
widths and spanned a period 100 ms before and after
the discharge of the reference unit (Fig. 1A). His-
tograms with a mean bin-count B4 were not analysed
further. The position and duration of the synchronous
peak was judged visually through the use of the cumu-
lative sum procedure (CUSUM; Ellaway (1978); dotted
vertical lines, Fig. 1A). The significance of synchronous
peaks in the cross-correlogram was assessed using the
method described by Wiegner and Wierzbicka (1987).
A standard peak duration of 11 ms, centred at time
zero, was used for quantification of the strength of
synchrony in the MU pair if no significant peak from
the cross-correlogram was identified. The strength of
MU short-term synchronization was quantified using
the synchronization index CIS (common input
strength), which is the number of synchronous dis-
charges in the cross-correlogram in excess of chance
(dark area, Fig. 1A) divided by the duration of the trial,
i.e. the frequency of extra synchronous discharges in
the MU pair (Nordstrom et al., 1992).
Data are presented as mean9 S.E. unless otherwise
stated. Linear regression was used to assess the rela-
tionships between estimates of MU synchronization
obtained with the two methods. Statistical significance
was regarded as PB0.05.
3. Results
The extent of MU synchrony in FDI was assessed
using the SEMG synchronization index for 189 MUs
from 28 FDI muscles in 16 subjects. MU synchroniza-
tion was also assessed from cross-correlation of dis-
charge of 498 MU pairs in the same muscles, which
included the 189 MUs used for the SEMG analysis.
The relationship between the estimates of MU syn-
chrony in FDI obtained with the two techniques is
shown for 28 muscles in Fig. 2. The mean strength of
MU synchronization in FDI muscles estimated using
the SEMG method varied over a 10-fold range (range
0.12 –1.15), with an average of seven (range 2–15)
reference MUs per muscle used to calculate the mean
SEMG synchronization index. The mean strength of
MU synchronization (CIS) using cross-correlation of
MU discharge varied over a 200-fold range (range
0.005–1.03 extra counts s − 1), with an average of 18
(range 7–30) MU pairs per muscle used to calculate the
mean synchronization index CIS. With both techniques
the estimates of the mean strength of MU synchroniza-
tion varied over a large range in different muscles. For
the 28 muscles, however, linear regression revealed no
significant correlation between the two measures of
MU synchronization (r 2 =0.04, Fig. 2). The two meth-
ods clearly do not provide equivalent estimates of the
overall strength of MU synchrony in FDI.
Fig. 2. Relationship between the estimates of the strength of motor
unit synchronization in FDI obtained using the surface electromyo-
gram (SEMG synchrony index) and the cross-correlogram (synchrony
index CIS). Data from 28 FDI muscles showing the mean SEMG
synchrony index for each muscle (mean seven reference MUs per
muscle, range 2 – 15) plotted against the mean MU synchrony index
CIS (mean 18 MU pairs per muscle, range 7 – 30) for the correspond-
ing muscle. Linear regression revealed no significant correlation be-
tween these two estimates of MU synchrony in FDI (r 2 =0.04).
Mean (9 S.D.) duration of MU synchrony in the 28
FDI muscles was 15.79 3.2 ms when estimated using
the 291 cross-correlograms with significant central
peaks (an average of 10 cross-correlograms per muscle,
range 1–23). The mean duration of MU synchrony in
these FDI muscles estimated with the SEMG method
was 16.8 93.4 ms (189 MUs). The two estimates of the
mean duration of short-term synchrony in these mus-
cles were not significantly different (paired t-test; P\
0.05, n= 28). There was no significant correlation
between the estimates of synchronous peak duration
obtained by the two methods (r 2 = 0.095, n= 28, Fig.
3).
The estimates of overall synchrony in FDI muscles
reported in Figs. 2 and 3 were obtained using a rela-
tively large sample of MUs for studies of this type (an
average per muscle of seven reference MUs for the
SEMG index and 18 MU pairs for the cross-correlo-
grams). Nevertheless, in view of the poor correlation
between the two estimates of synchrony in FDI, it is
necessary to consider whether the MU sample in each
muscle was sufficiently large to provide a reliable esti-
mate of the overall extent of MU synchrony in the
muscle. If some FDI MUs have consistently very weak
or strong synchronization with many other MUs, such
a non-uniformity in the distribution of synchronization
within the MU population would increase the possibil-
ity that under-sampling could produce errors in the
estimates of global synchronization in the muscles
which may have obscured the expected positive correla-
tion between the two synchronization indices in Fig. 2.
Sampling error is more likely to be an issue for the
cross-correlogram analyses, because with the SEMG
 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
technique all active MUs contribute to the SEMG
signals used to calculate the synchronization index for
each reference MU.
While it is common to encounter a MU pair with
unusually weak, or strong, synchronization in any mod-
erate-sized sample of the MU population of a muscle, it
is not known whether this represents a chance varia-
tion, or a consistent tendency for one or both of these
MUs to exhibit unusually weak or strong synchroniza-
tion with many or all MUs in the muscle. To address
this question, it is necessary to consider the strength of
synchrony exhibited by a single MU when paired with
many other concurrently active MUs, and compare this
with the overall extent of synchrony revealed in cross-
correlograms from a large sample of other MUs in that
muscle. Sufficient data for such a comparison are
difficult to obtain, but in our sample there were 13
MUs (from 10 FDI muscles) that were paired for
cross-correlation with at least five other concurrently
active MUs. For each of these 13 selected MUs, we
calculated a measure of that MU’s synchronization
with other concurrently active MUs by pooling CIS
values from all available cross-correlograms for which
that MU acted as the reference (this ranged between 5
and 13, with an average of 7). This was compared with
the mean CIS obtained by pooling data from cross-cor-
relograms involving all other MU pairs in that FDI
muscle (i.e. excluding pairs involving the selected MU).
This ranged between 9 and 23 cross-correlogram pairs,
with an average of 14. The results of this analysis for
the 13 MUs are summarised in Fig. 4.
Fig. 4 shows for each of 13 selected MUs, the mean
synchronization index CIS for cross-correlograms with
that unit as the reference, plotted against the mean CIS
Fig. 3. Relationship between the duration of the central synchronous
peak in FDI obtained with the two methods. Data for 28 FDI
muscles showing the mean duration of the central synchronous peak
using the SEMG method plotted against the mean duration of the
significant synchronous peaks in the cross-correlograms (total 291
MU pairs) obtained from MUs in these muscles. Linear regression
revealed no significant correlation between these two estimates of the
duration of MU synchrony in FDI (r 2 = 0.05).
51
Fig. 4. Comparison using cross-correlation of the strength of MU
synchrony for a restricted sample of MUs and a global estimate of
MU synchrony in FDI. Abscissa: data from 13 selected FDI MUs
which were cross-correlated with at least five other concurrently
active MUs. The mean synchrony index CIS from the restricted
sample of cross-correlograms involving the selected reference MU is
plotted against the mean synchrony index CIS from all remaining
cross-correlograms for that FDI muscle (ordinate). The restricted
sample was a reasonably good predictor of the global estimate of FDI
MU synchrony (linear regression line shown: r =0.86; P B0.001).
Filled symbols show the two selected MUs with significantly higher
synchronization in their cross-correlograms than the global estimate
for that FDI muscle.
obtained by pooling data from all other MU pairs in
the muscle. Linear regression of these data revealed a
significant positive correlation (r= 0.86; PB0.001).
This indicates that data from a single reference MU
obtained using five or more cross-correlograms can be
used to provide a reasonable estimate of overall MU
synchrony in the FDI muscle. In two of 13 cases (Fig.
4, filled circles), however, the mean synchronization of
the selected reference unit with other MUs was signifi-
cantly different from the overall estimate of MU syn-
chrony obtained from cross-correlation of all other
available MU pairings (t-tests with Bonferroni correc-
tion for multiple contrasts). In one case, the mean9
S.E. (n) CIS of cross-correlograms involving the
selected MU was 0.7590.18 (7) versus 0.319 0.05 (23)
for all other MU pairings in that muscle (PB 0.05). In
the second case, mean CIS values were 0.46 9 0.05 (7)
for pairings with the selected MU versus 0.1590.02
(17) for all other MU pairings in the muscle (PB
0.001). These data suggest that a minority of MUs are
more highly synchronized than the rest of the motoneu-
ron pool, although in general MU synchrony is rela-
tively uniformly distributed amongst the MU pool in
FDI.
52 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
The preceding analyses (Fig. 4) have established that
the overall extent of MU synchrony in FDI when
quantified by pooling cross-correlation data from many
MU pairs is reasonably well estimated by sampling one
reference MU that is cross-correlated with at least five
other MUs. For 11 of this sample of 13 MUs, the
pattern of MU synchrony was consistent with the be-
haviour of a larger sample of MUs in the muscle. If the
SEMG index is a reliable estimate of the overall extent
of synchrony between the reference unit and other
concurrently active MUs in the muscle, one would
expect a significant positive correlation between the
mean CIS estimated from cross-correlograms involving
the 13 MUs selected for the Fig. 4 analysis, and the
SEMG synchronization index calculated for the same
13 MUs. These data are shown in Fig. 5. There was no
significant correlation between the two synchrony mea-
sures obtained using these 13 MUs as the reference
(r = 0.24; P\0.05). When all available data were used
for the regression analysis (184 MUs for which a
SEMG synchrony index was calculated, and the MU
was paired with at least one other MU for an estimate
of synchrony from the cross-correlogram) the relation-
ship between the SEMG synchrony index and mean
CIS was not significant (r =0.14, P =0.06).
An additional comparison was used to assess the
relationship between the two methods of quantifying
MU synchrony for a population of MUs. We selected
Fig. 5. Relationship between estimates of MU synchrony obtained
with SEMG and cross-correlation techniques for 13 FDI MUs. Data
from 13 FDI MUs selected because they were cross-correlated with at
least five other concurrently active MUs. For each MU, the SEMG
synchrony index is plotted against the mean synchronization index
CIS from all cross-correlograms involving that MU. There was no
significant correlation between the two estimates of MU synchrony
obtained using these 13 reference MUs.
pairs of MUs in FDI that exhibited large differences in
the SEMG synchrony index (0.3 or greater), and for
which there were at least two cross-correlograms avail-
able for each MU of the pair (sum total of cross-correl-
ograms involving the two units ranged from 5 to 12,
with a mean of 8). These selection criteria resulted in 20
MU pairs for the analysis. If the two methods of
quantifying MU synchrony were related, it was ex-
pected that the MU of the pair with the higher SEMG
synchrony index would tend to have the higher mean
synchronization in the cross-correlograms. This pattern
was observed in only nine of 20 comparisons. In the
remaining 11 pairs, the MU with the higher SEMG
synchrony index had the lower mean CIS of the pair.
This distribution of mean CIS differences for the 20
MU pairs was not significantly different from that
expected due to chance (sign test; P = 0.83), so once
again for a group of MUs it was not possible to
demonstrate a relationship between the estimates of
MU synchrony obtained with the SEMG and cross-cor-
relation methods.
4. Discussion
Implicit in the use of the SEMG synchronization
index is the assumption that it provides similar infor-
mation on the extent of MU synchronization in a
muscle as the cross-correlation technique, yet this
proposition had not previously been tested quantita-
tively. Fig. 2 shows that estimates of the strength of
MU synchrony in FDI muscles obtained using the two
techniques were not related in any meaningful way,
despite a wide range of values for both measures in the
28 FDI muscles studied and considerable overlap in the
identity of MUs used to quantify synchrony with each
measure. This suggests that either one or both of the
techniques are not providing an accurate measure of
MU synchrony in the muscle. Both techniques seem to
be reflecting short-term synchronization, as the mean
duration of the synchronous peaks was brief (approxi-
mately 16 ms), and estimates of peak duration obtained
with the two techniques were not significantly different
(Fig. 3).
The discrepancy between the estimates of MU syn-
chronization obtained with the two methods is nicely
illustrated by the data from two MUs shown in Fig. 1.
For this MU pair, cross-correlation revealed a high
degree of synchronization (Fig. 1A, CIS of 1.21). The
SEMG synchronization index for one of these units was
rather low (Fig. 1B, 0.31) and for the other unit it was
rather high (Fig. 1C, 1.03). One interpretation of these
data is that the discharge of this pair of MUs is highly
synchronized, but the MU with the low SEMG syn-
chronization index (Fig. 1B) has weak synchrony with
the majority of other active MUs in FDI, while the unit
 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
with the high SEMG index (Fig. 1C) is consistently
strongly synchronized with other MUs. If MU syn-
chrony is non-uniformly distributed within FDI, this
would raise some concerns about the sample size
needed to obtain a good estimate of the overall strength
of MU synchrony in a muscle. An alternative explana-
tion is that the high SEMG synchrony index for the
MU depicted in Fig. 1C is artifactual, and arises from
technical limitations of the SEMG technique (cf. Yue et
al., 1995). One problem seen here is that this MU’s
representation in the SEMG signal is small, and barely
sufficient to exceed the baseline average rectified EMG
level (compare the size of the diagonally hatched areas
in Fig. 1B and C). This poor signal-to-noise ratio will
lead to a high value for the synchronization index, even
with a constant contribution to the peak in the rectified
EMG average from synchronized MUs (Yue et al.,
1995).
4.1. Sampling issues and the uniformity of MU
synchronization for different MUs in a muscle
Both techniques rely on sampling a proportion of
active MUs in FDI to estimate the overall extent of
MU synchrony in the muscle. The human FDI has
about 120 MUs (Feinstein et al., 1955). All MUs
included in the present study had recruitment force
thresholds B4 N, which is equivalent to about 10–15%
of FDI maximal voluntary force. It is difficult to esti-
mate how many MUs are active in FDI at these low
force levels, but it is likely to be a small fraction of the
total, as the abduction force during the trials was
commonly between 0.5 and 1 N. The mean number of
reference MUs per muscle used for the SEMG syn-
chrony analyses was seven, and the mean number of
cross-correlograms analysed per muscle was 18. The
MU sample was considered to be reasonable for the
SEMG technique, in view of the fact that all active
MUs potentially contribute to the SEMG synchrony
index obtained for each reference MU. The number of
MU pairs used for the cross-correlogram analyses is a
more limited proportion of the possible combinations
of active MUs, and for this reason we looked for
evidence on the sample size needed to provide a reason-
able estimate of overall MU synchronization using the
cross-correlation technique. Lower limits on the sample
size required for a valid estimate of overall MU syn-
chrony in a muscle, and the issue of the uniformity of
the distribution of MU synchrony within the MU pop-
ulation, have not been explored previously. This is
presumably because sufficient data for the comparisons
are difficult to obtain.
In 11 of 13 (85%) cases in which an MU was paired
with at least five other concurrently active MUs for
cross-correlation analysis, the mean synchronization in-
dex for the restricted sample involving a common refer-
53
ence MU did not differ significantly from the mean
synchronization index obtained by pooling cross-corre-
lation data from all other available MU pairings in that
muscle (Fig. 4). There was a significant positive correla-
tion (r= 0.86) between the two synchrony measures
obtained from the restricted and larger sample of cross-
correlograms (Fig. 4). However, two of 13 MUs studied
were significantly more synchronized with other MUs
than the global average obtained from all other MU
pairings in the muscle. This indicates that there is not a
completely uniform distribution or effectiveness of
shared, branched-axon inputs to all motoneurons in the
pool; a minority (approximately 15% in our sample) of
motoneurons share more common inputs than is the
norm for their motoneuron pool. This finding supports
the view that the previously observed large variation in
the strength of MU synchrony for individual MU pairs
in a single FDI muscle (Bremner et al., 1991; Nord-
strom et al., 1992) does not arise by chance.
The Fig. 4 analysis indicates that a reasonable esti-
mate of the overall extent of MU synchrony in FDI can
be obtained by cross-correlating the discharge of one
reference MU with as few as five other MUs. This
procedure provided an estimate of MU synchrony that
was positively correlated with the global estimate of
synchrony, and not significantly different from it in
85% of comparisons. As the average number of MU
pairs/muscle used to provide a global estimate of FDI
MU synchrony from the cross-correlograms was 18 for
the Fig. 2 analysis (minimum seven), and these did not
include the same reference MU in every pairing, we feel
justified to conclude that the number and diversity of
MU pairs sampled per muscle was sufficient to provide
a reliable global estimate of the strength of FDI MU
synchronization with the cross-correlation technique.
The 13 MUs that could be cross-correlated with five
or more MUs were used for a direct comparison of the
synchrony measures obtained with the SEMG and
cross-correlation methods (Fig. 5). Even in this ‘best-
case’ comparison, the synchrony measures obtained
with the two techniques were not significantly corre-
lated. Similarly, a comparison using 20 MU pairs with
large differences in the SEMG synchrony index re-
vealed that the MU of the pair with the higher SEMG
synchrony index had no significant tendency to exhibit
higher mean synchrony in the cross-correlograms with
other MUs.
Our data consistently show that measures of MU
synchrony obtained using the SEMG and cross-correlo-
gram techniques do not bear any useful relationship to
one another. Our evidence suggests that these disparate
findings are not due to under-sampling of the MU
population, despite a tendency for non-uniform syn-
chronization within a sub-set of the MU population in
FDI. We now turn to consideration of methodological
limitations of the SEMG method.
54 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
4.2. Technical limitations of the surface EMG method
for quantifying MU synchrony
The SEMG technique is an indirect estimate of MU
synchrony, with the complication that contributions of
single MUs to the SEMG are not equal (depending on
the size and number of muscle fibres, and their location
and orientation with respect to the recording elec-
trodes). An important limitation of the SEMG method
is that, on theoretical grounds (Milner-Brown et al.,
1973), slightly positive synchronization ratios would be
expected under asynchronous conditions. This is due to
the partial summation of the rectified waveform and the
background EMG activity, and is termed the rectifica-
tion artifact. This problem was recognised originally by
Milner-Brown et al. (1973) who used a theoretical
approach to calculate the contribution of the rectifica-
tion artifact that accounted for changes in the signal-to-
noise ratio. When the rectification artifact is calculated
and subtracted from the synchronization index, the
SEMG technique has been shown to provide a sensitive
measure of the level of MU synchronization within a
population of neurons (Roscoe et al., 1985). However,
due to the extra data processing required to calculate
the rectification artifact, Milner-Brown et al. (1975)
chose to simplify the method by assuming that the
rectification artifact was fixed, and was independent of
the signal-to-noise ratio. This simplified SEMG method
was used by these authors to show an increase in MU
synchronization with strength-training (Milner-Brown
et al., 1975), a result that is widely cited in the literature
as a neural adaptation to strength training (Sale, 1987).
Technical problems associated with the SEMG
method have recently been highlighted by Yue et al.
(1995). Using experimental data from FDI MUs made
to discharge under low- and high-rate conditions, Yue
et al. (1995) showed that the SEMG synchrony index
was significantly positively correlated with the level of
EMG activity in the muscle. This was attributed to the
recruitment of additional synchronized MUs at higher
levels of muscle activation. Furthermore, using a com-
puter simulation of 100 FDI MUs discharging asyn-
chronously, Yue et al. (1995) reported that the
rectification artifact contributed from 0 to 0.15 to the
SEMG synchronization index, and this figure varied
non-linearly with the signal-to-noise ratio (defined in
this case as the ratio of the peak amplitude of the
rectified average relative to the average background
EMG level). The signal-to-noise ratio was reduced at
higher levels of background EMG activity, and as the
signal-to-noise ratio decreased the accuracy of the
SEMG index as an estimate of MU synchronization
declined.
Two recommendations were made by Yue et al.
(1995) to minimise errors when using the SEMG
method to estimate MU synchrony, but they predicted
that these were likely to be so difficult to achieve under
normal experimental conditions that use of the method
to accurately assess MU synchrony was not practical.
The first was that the SEMG technique only be used
when the rectification artifact was negligible, which in
their simulation occurred when the peak rectified EMG
signal in the average was more than three times as large
as the background EMG level. In the present study,
only 12 of 189 MUs (range 1–6 reference MUs per
muscle) in six muscles satisfied this rather stringent
criterion. For these MUs, there was no significant
relationship between their SEMG synchrony index and
the mean synchronization index CIS from all cross-cor-
relograms (ranging 1–3) involving the MU (r=0.52,
P\ 0.05). The rectified EMG peak was more than
twice as large as the background EMG level for only 43
MUs in the present study. For these 43 MUs there was
also no significant relationship between SEMG syn-
chrony index and the mean synchronization index CIS
from all cross-correlograms (ranging 1–10) involving
the MU (r= 0.17, P\ 0.05).
The second recommendation of Yue et al. (1995) was
that comparisons between different muscles should only
be made for contractions involving a quite narrow
range of forces. This was because of the dependence of
the SEMG synchronization index on the level of muscle
excitation (presumably because recruitment influences
the number of active MUs available to contribute to
the global synchrony measure). In their study of 56
FDI MUs tested at two different levels of muscle
activation, Yue et al. (1995) found that the surface
synchrony index was 31% higher when the average
muscle contraction force increased from about 4 to 7%
of MVC. Most of the index finger abduction forces at
which FDI MUs were studied in the present study were
in the range 0.1–5% MVC, with no contraction over
12% MVC. It is likely that differences in the number of
active MUs over the force range at which different
MUs were tested were responsible for considerable
variation in the SEMG synchronization indices in the
present study.
In summary, there is a poor correlation between
global estimates of MU synchronization in FDI mus-
cles using MU cross-correlation and SEMG methods.
Cross-correlation of MU discharge times is a more
direct, and therefore more reliable measure of MU
synchronization, provided that sufficient MU pairs are
included in the analysis. Although the tendency for
synchronous discharge was higher than average for a
minority (15%) of MUs, we have presented evidence
that our sample size was sufficient to obtain a reliable
global estimate of MU synchrony in FDI using cross-
correlation. The most likely explanation for the poor
correlation between the estimates of MU synchrony
obtained with the two methods is technical limitations
of the SEMG method (Yue et al., 1995), which are
 J.G. Semmler, M.A. Nordstrom / Journal of Neuroscience Methods 90 (1999) 47–55
difficult to overcome under normal experimental condi-
tions. We conclude that the simplified SEMG method
(Milner-Brown et al., 1973, 1975) is not reliable for
quantitative comparisons of MU synchrony between
muscles and subjects.
Acknowledgements
This work formed part of the PhD studies of JGS,
who was supported by a University of Adelaide Post-
graduate Research Scholarship. MAN was an R.D.
Wright Fellow of the NH&MRC of Australia.
References
Bremner FD, Baker JR, Stephens JA. Variation in the degree of
synchronization exhibited by motor units lying in different finger
muscles in man. J Physiol (Lond) 1991;432:381–99.
Datta AK, Stephens JA. Synchronization of motor unit activity
during voluntary contraction in man. J Physiol (Lond)
1990;422:397 – 419.
Datta AK, Farmer SF, Stephens JA. Central nervous pathways
underlying synchronization of human motor unit firing studied
during voluntary contractions. J Physiol (Lond) 1991;432:401 – 25.
Ellaway PH. Cumulative sum technique and its application to the
analysis of peristimulus time histograms. Electroencephalogr Clin
Neurophysiol 1978;45:302–4.
Farmer SF, Swash M, Ingram DA, Stephens JA. Changes in motor
unit synchronization following central nervous lesions in man. J
Physiol (Lond) 1993;463:83–105.
55
Feinstein B, Lindegard B, Nyman E, Wohlfart G. Morphological
studies of motor units in normal human muscles. Acta Anat
1955;23:127 – 42.
Milner-Brown HS, Stein RB, Yemm R. The contractile properties of
human motor units during voluntary isometric contractions. J
Physiol (Lond) 1973;228:285 – 306.
Milner-Brown HS, Stein RB, Lee RG. Synchronization of human
motor units: possible roles of exercise and supraspinal reflexes.
Electroencephalogr Clin Neurophysiol 1975;38:245 – 54.
Moore GP, Perkel DH, Segundo JP. Statistical analysis and func-
tional interpretation of neuronal spike data. Ann Rev Physiol
1966;28:493 – 522.
Nordstrom MA, Fuglevand AJ, Enoka RM. Estimating the strength
of common input to human motoneurons from the cross-correlo-
gram. J Physiol (Lond) 1992;453:547 – 74.
Roscoe DD, Hamm TM, Reinking RM, Stuart DG. Detection of
synchrony in the discharge of a population of neurons. II. Imple-
mentation and sensitivity of a synchronization index. J Neurosci
Methods 1985;13:51 – 64.
Sale DG. Influence of exercise and training on motor unit activation.
Exerc Sport Sci Rev 1987;15:95 – 151.
Sears TA, Stagg D. Short-term synchronization of intercostal mo-
toneurone activity. J Physiol (Lond) 1976;263:357 – 81.
Semmler JG, Nordstrom MA. Influence of handedness on motor unit
discharge properties and force tremor. Exp Brain Res
1995;104:115 – 25.
Semmler JG, Nordstrom MA. Motor unit discharge and force tremor
in skill- and strength-trained individuals. Exp Brain Res
1998;119:27 – 38.
Wiegner AW, Wierzbicka MM. A method for assessing significance
of peaks in cross-correlation histograms. J Neurosci Methods
1987;22:125 – 31.
Yue G, Fuglevand AJ, Nordstrom MA, Enoka RM. Limitations of
the surface electromyography technique for estimating motor unit
synchronization. Biol Cybern 1995;73:223 – 33.