Clinical Neurophysiology 121 (2010) 1616–1623

Contents lists available at ScienceDirect

Clinical Neurophysiology
journal homepage: www.elsevier.com/locate/clinph

Invited review

Decoding the neural drive to muscles from the surface electromyogram

Dario Farina a,*, Aleš Holobar b,c, Roberto Merletti b, Roger M. Enoka d
a
Center for Sensory-Motor Interaction (SMI), Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
b
Laboratory for Engineering of the Neuromuscular System, Department of Electronics, Politecnico di Torino, Torino, Italy
c
Faculty of Electrical Engineering and Computer Science, University of Maribor, Maribor, Slovenia
d
Department of Integrative Physiology, University of Colorado at Boulder, CO, USA

a r t i c l e i n f o a b s t r a c t

Article history: This brief review discusses the methods used to estimate the neural drive to muscles from the surface
Accepted 22 October 2009 electromyogram (EMG). Surface EMG has been classically used to infer the neural activation of muscle
Available online 4 May 2010 by associating its amplitude with the number of action potentials discharged by a population of motor
neurons. Although this approach is valuable in some applications, the amplitude of the surface EMG is
Keywords: only a crude indicator of the neural drive to muscle. More advanced methods are now available to esti-
EMG
mate the neural drive to muscle from the surface EMG. These approaches identify the discharge times of a
Amplitude cancellation
Motor unit
few motor units by decomposing the EMG signal to determine the relative changes in neural activation.
Decomposition This approach is reliable in several conditions and muscles for isometric contractions of moderate force,
Spike train but is limited to the few superficial units that can be identified in the recordings.
Ó 2010 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights
reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1616
2. Amplitude of the surface EMG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1617
3. Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1619
4. Representativeness and accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1621
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1622
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1622
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1622

1. Introduction associated with the magnitude of the motor command discharged

by the motor neuron pool.
The central nervous system controls the force generated by a The electrical activity in muscle can be recorded with electrodes
muscle with recruitment of motor units and adjustments in their placed over the skin. The resulting surface electromyogram (EMG)
discharge rates (Adrian and Bronk, 1929). The number of action is the sum of the action potentials generated by the motor units
potentials discharged by the motor neurons innervating the mus- and filtered by the volume conductor. The shape of each motor unit
cle constitutes the neural drive from the spinal cord to the muscle. action potential depends on the number and anatomical character-
Due to a high safety margin at the nerve–muscle synapse, action istics of the innervated muscle fibers, and on the properties of the
potentials discharged by the motor neurons usually emerge as ac- recording electrodes (Farina et al., 2002). The magnitude of the
tion potentials that propagate along the muscle fibers. Thus, the neural drive to muscle can be quantified from the surface EMG
ensemble of electrical activity generated by a muscle is strongly by identifying the number of motor unit action potentials gener-
ated per time unit.
Although the amplitude of the surface EMG is often used to esti-
mate the magnitude of the neural activation sent to muscle, there
* Corresponding author. Address: Center for Sensory-Motor Interaction (SMI),
are some limitations to this interpretation. First, the size of a motor
Department of Health Science and Technology, Aalborg University, Fredrik Bajers
Vej 7 D-3, DK-9220 Aalborg, Denmark. Tel.: +45 96358821; fax: +45 98154008. unit action potential at the surface of the skin is only partly related
E-mail address: [email protected] (D. Farina). to motor neuron size (Keenan et al., 2006). Second, the relative

1388-2457/$36.00 Ó 2010 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.clinph.2009.10.040
 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623 1617

contribution of an action potential to the amplitude of the EMG

signal may differ across conditions (Dimitrova and Dimitrov,
2003). To circumvent these limitations, an alternative approach
is to use the times of occurrence (discharge times) of the action
potentials of individual motor units, identified by surface EMG
decomposition, to estimate the motor output from the spinal cord.
This brief review discusses the methods used to estimate the
neural drive to muscles from the surface EMG. As detailed issues
related to classic approaches were reviewed previously (Farina
et al., 2004a), this review focuses on recent developments in inter-
preting surface EMG amplitude and in decomposing the surface
EMG to detect discharge times.

2. Amplitude of the surface EMG

The amplitude of the surface EMG is usually estimated as the

standard deviation of the signal (Clancy and Hogan, 1994). The
average rectified value (ARV) and the root mean square (RMS) val-
ues of the surface EMG are optimal estimators of amplitude when
the EMG signal follows Laplacian and Gaussian distributions,
respectively (Clancy and Hogan, 1999). These two estimators are
equivalent in practical applications (Clancy et al., 2002). The use
of signal amplitude as an indicator of the neural activation has
two major drawbacks. First, the size of surface action potentials
differs among motor units and across conditions, which alters
the association between number of motor neuron discharges and
signal amplitude. Second, the sum of the surface action potentials
is less than the sum of the amplitudes of the individual potentials,
a phenomenon that has been termed ‘‘amplitude cancellation”
Fig. 1. (A) Surface EMG signal detected from the abductor digiti minimi muscle of a
(Day and Hulliger, 2001; Keenan et al., 2005), which results in healthy man during an isometric contraction of 10 s duration at 10% of the maximal
the amplitude of the surface EMG underestimating the magnitude force. Intramuscular EMG signals were concurrently recorded with two pairs of
of the motor output from the spinal cord and disrupts the associa- wire electrodes and single motor unit action potentials were identified by
tion between changes in the two signals. decomposition (McGill et al. 2005). (B) Smoothed discharge rates were obtained
by filtering the point process representing the inverse of the interspike intervals
The variability in action potential size among motor units and
(high-pass filtered with cut-off frequency 0.75 Hz and smoothed with Hann
across conditions has been analyzed in several studies (Farina window of 400 ms duration) of the motor units identified from the intramuscular
et al., 2002; Dimitrova and Dimitrov, 2003; Dimitrov et al., recordings (n = 9 motor units). The neural activation was estimated by counting the
2008). It seems that the size of a surface action potential is only total number of action potentials discharged by the 9 units in each 400 ms interval.
moderately associated with the size of the motor unit (Keenan (C) Comparison between the abduction force exerted by the little finger during the
contraction (grey line) and the neural activation estimated from the count of the
et al., 2006) and that, conversely, the size depends on factors that number of action potentials in each 400 ms interval (black line) after removing the
are difficult to control or limit in experimental conditions (e.g., the offset in the two signals. (D) Comparison between the estimated neural activation
thickness of the subcutaneous layer; Dimitrov et al., 2002). Thus, from panel B (black line) and the amplitude of the surface EMG (grey line). The
the set of surface action potentials that represent the activity of a amplitude of the surface EMG was estimated as the root mean square value in the
same intervals used for estimating the neural activation (400 ms in duration). au,
population of motor neurons has a broad distribution of sizes,
Arbitrary units; pps, pulses per second.
which differs among subjects, conditions, and muscles. The abso-
lute EMG amplitude (without normalization), therefore, provides
a poor index of the neural drive to the muscle and absolute com- the surface EMG provides only a crude measure of the changes in
parisons are not appropriate (Dimitrova and Dimitrov, 2003). neural activation.
Rather, the surface EMG amplitude should be normalized to a value The amount of amplitude cancellation experienced by individ-
obtained in a reference contraction, such as a maximal voluntary ual action potentials can be predicted analytically (Farina et al.,
contraction (Keenan et al., 2005; Yang and Winter, 1984). 2008a). It is greater for small potentials and increases as the ampli-
Due to amplitude cancellation of motor unit action potentials in tude of the interference signal increases (Farina et al., 2008a). Be-
surface EMG recordings, the relation between surface EMG ampli- cause low-threshold motor units tend to produce smaller surface
tude and number of action potentials per time unit would not be action potentials than high-threshold units (Keenan et al., 2006),
linear even if all the action potentials had the same size (Farina it is the surface action potentials of the low-threshold motor units
et al., 2004a; Keenan et al., 2005). Amplitude cancellation nega- that are reduced more by amplitude cancellation, especially for
tively impacts the association between fluctuations in surface high levels of muscle activity. As a consequence of this effect, the
EMG amplitude and muscle force. Fig. 1 shows that small changes surface EMG amplitude can be relatively insensitive to changes
in the neural drive to the abductor digiti minimi muscle at low in the activity of low-threshold motor units. For example, Mottram
contraction forces, as estimated by counting the number of action et al. (2005) found that the discharge rate of low-threshold motor
potentials discharged by nine motor units per unit time (Fig. 1B), units decreased more rapidly during one type of fatiguing contrac-
were strongly correlated with the exerted force (R = 0.88; Fig. 1C) tion compared with another type, whereas the change in the
but not with the surface EMG amplitude (R = 0.21; Fig. 1D). Also, amplitude of the surface EMG was similar during the two tasks.
there was a weak correlation between surface EMG amplitude Similarly, indexes of neural drive to the muscle based on EMG
and force (R = 0.28). This example indicates that the amplitude of amplitude, such as the neuromuscular efficiency (Deschenes
1618 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623

et al., 2002), are largely biased toward changes in the activity of motor units changed in response to the simulated accumulation
high-threshold motor units due to the influence of amplitude of metabolites, according to a model of the intra- and extra-cellular
cancellation. compartments (Dideriksen et al., 2009). Surface EMG action poten-
The amount of amplitude cancellation may vary during fatigu- tials were generated by a validated model (Farina et al., 2004b).
ing contractions due to changes in the interference EMG amplitude The time to task failure in these simulations was 293 s for 30%
and the shapes of the motor unit action potentials. When an indi- MVC and 79 s for 60% MVC (Fig. 2A).
vidual sustains a submaximal contraction, the amplitude of the The number of action potentials discharged each second by all
surface EMG usually increases due to the recruitment of additional active motor units (expressed as % of the number during an
motor units (Garland et al., 1994; Person and Kudina, 1972; Riley MVC) increased from 28% at the beginning of the contraction to
et al., 2008), the decrease in muscle fiber conduction velocities 61% at the end (30% MVC force) and from 52% to 62% (60% MVC
(Bigland-Ritchie et al., 1981; Merletti et al., 1990), and changes force) (Fig. 2C). This increase was due to the recruitment of addi-
in the shapes of the intracellular action potentials (Dimitrova and tional motor units during the contraction. Although the number
Dimitrov, 2002; Hanson and Persson, 1971). The adjustments that of action potentials per time unit was greater at the beginning of
occur during fatiguing contractions change the distribution of ac- the stronger contraction (52% vs. 28%), it was similar at task failure
tion potential size and the amount of amplitude cancellation, for the two tasks. The surface EMG amplitude (RMS) at task failure
which alters the relation between the neural activation of muscle was 62% and 88% of the MVC value for the 30% MVC and 60% MVC
and surface EMG amplitude, even when this is normalized. contractions, respectively (Fig. 2D), despite a similar number of ac-
Fig. 2 shows a simulation analysis that exemplifies the limita- tion potentials discharged per second. This discrepancy indicates
tions of surface EMG amplitude as a measure of neural activation that the relative changes in surface EMG amplitude during the sim-
during sustained contractions. In this example, a population of mo- ulated contraction failed to represent some significant changes in
tor neurons was activated (Fuglevand et al., 1993) to sustain con- the neural drive. The dissociation between the changes in EMG
tractions at 30% and 60% of the maximal voluntary contraction amplitude and number of discharged action potentials were partly
(MVC) force until task failure, which corresponded to a decrease due to differences in the amount of amplitude cancellation
in simulated force by 2% MVC force for more than 2 s. The contrac- (Fig. 2E). Amplitude cancellation was similar at the beginning of
tile properties, recruitment, and discharge characteristics of the the tasks (30% and 31%, respectively) but increased more for the

Fig. 2. (A) Simulated forces during contractions at 30% and 60% of the maximal voluntary contraction (MVC) force. The open (60% MVC) and filled (30% MVC) circles indicate
the times of task failure for the two contractions, which was defined at the time when the force decreased by 2% MVC force for 2 s. The simulations represent contractions
with the first dorsal interosseus muscle. The number of motor units in the muscle was 120 and the upper limit of recruitment in the absence of fatigue was 60% MVC force.
The other model parameters at the beginning of the tasks were similar to those described by Keenan et al. (2005); however, the parameters changed over time to simulate the
adjustments that occur during fatiguing contractions (Dideriksen et al., 2009). (B) The simulated surface EMG signals during the two contractions. Motor unit action
potentials were generated with the EMG model proposed by Farina et al. (2004b). (C) The total number of discharges of all active motor units, expressed as a percentage of the
number of discharges during an MVC contraction for the 30% MVC (filled circles) and the 60% MVC (open circles) tasks. The number of discharges was computed for time
intervals corresponding to 5% of the time to failure for each task. (D) The root mean square (RMS) value of the surface EMG signal computed for the same time intervals as in
(C). (E) The amount of amplitude cancellation in the simulated surface EMG signals for the same time intervals as in (C) and (D). Amplitude cancellation was computed as the
difference in the amplitudes of the rectified and summed action potential trains (Keenan et al., 2005), which is only possible in simulated conditions. au, Arbitrary units; MU,
motor unit.
 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623 1619

30% MVC (38% at task failure) than for the 60% MVC task (32% at (Holobar et al., 2009; see Merletti et al. (2008) for a review).
task failure). Although different trends in EMG amplitude can be Decomposition involves compensating for the shapes of the action
obtained by choosing other model parameters, this example indi- potentials and identifying the underlying trains of motor unit ac-
cates that changes in surface EMG amplitude provide a poor index tion potentials (Holobar and Zazula, 2007). Contrary to simpler ap-
of the modulation in neural activation during sustained proaches (Kallenberg and Hermens, 2006), the methodology
contractions. requires distinguishing the action potentials of different motor
units, which is facilitated by recording the surface EMG from sev-
eral locations over the muscle (Farina et al., 2008b; Fig. 3). These
3. Decomposition
‘‘high-density surface EMG” systems (Blok et al., 2002; Merletti
et al., 2008) comprise several, closely spaced surface electrodes.
The main issue when using EMG amplitude to estimate the neu-
Fig. 3B shows the amplitude distribution of surface EMG signals
ral drive to muscles is the influence of differences in action poten-
recorded by 61 electrodes, from which 56 bipolar derivations were
tial shape on the composite signal. The use of EMG amplitude for
obtained, at one instant in time. The amplitude values were deter-
this purpose, however, assumes that the distribution of action po-
mined from the action potentials of the motor units that were ac-
tential shapes does not have a significant influence on the results,
tive at that instant. The amplitude distributions of the motor unit
but this is not correct. In contrast, identification of the times at
potentials that contributed to the interference signal at the indi-
which action potentials are discharged by individual motor units
cated instant were obtained with a decomposition method (Holo-
from the interference EMG allows more direct access to the neural
bar and Zazula, 2007). The sum of the motor unit amplitude
coding. The procedure to determine the discharge times is referred
distributions (Fig. 3C) corresponds to the amplitude of the interfer-
to as decomposition of the EMG signal (De Luca et al., 2006).
ence signal (Fig. 3B) with small error. These associations are valid
The classic approach to identifying motor unit action potentials
for all other time instants and allow unequivocal identification of
is based on intramuscular EMG signals (Adrian and Bronk, 1929) as
the motor unit discharge times as the initiation of the correspond-
they provide greater selectivity than surface EMG signals. Recent
ing action potentials (decomposition of the signal). Conversely, the
advances, however, have developed the option of decomposing
amplitude values of the interference signal recorded from only a
the surface EMG into its constituent motor unit action potentials

Fig. 3. (A) Surface EMG signals recorded from the abductor pollicis brevis muscle of a healthy man during an isometric contraction sustained at 10% of the maximal voluntary
contraction (MVC) force. The signals were detected as bipolar recordings from an electrode grid of five columns and 13 rows (with four missing electrodes at the corners; 56
bipolar recordings; 5 mm interelectrode distance), with the columns aligned along the fiber direction. (B) Amplitude of the multi-channel EMG recording for all electrode
locations at the instant indicated by the small dots and vertical lines in (A). Some of the recording locations are indicated by numbers corresponding to those in (A). The multi-
channel surface EMG signal was decomposed (Holobar and Zazula, 2007) to identify 18 motor units. Eight of the 18 motor units had action potentials at the instant indicated
in (A) and these contributed to the amplitude of the interference EMG signal at that instant. (C) The amplitude of the action potentials of the eight motor units at the instant
indicated in (A). The amplitude distribution of the interference signal shown in (B) corresponds closely to the sum of the amplitude distributions in (C). The colorbar is
common to all amplitude maps. MU, motor unit.
1620 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623

few electrode locations can be described by multiple combinations due to the high-density EMG recording (Farina et al., 2008b). Even
of motor unit action potentials, thus the decomposition would when action potentials of two different motor units have similar
have multiple solutions (Farina et al., 2008b). shapes on most of the multi-channel recordings (e.g., motor units
Among the methods developed for decomposing the surface 2 and 8 in Fig 4C), the small but consistent differences on some
EMG, the convolution kernel compensation (CKC) technique (Holo- of the other channels allow the decomposition of the multi-chan-
bar and Zazula, 2007) has been extensively validated with both nel EMG to discriminate between the two motor units. The exam-
simulated and experimental signals (Holobar et al., 2009). Fig. 4 ple shown in Fig. 4 presents the discharge times of motor unit
shows an example of this procedure for experimental signals re- action potentials identified during a ramp contraction so that the
corded with a high-density surface EMG system from the biceps recruitment thresholds of the motor units could be estimated
brachii muscle. Each motor unit is represented by its template ac- (Farina et al., 2009).
tion potential in terms of its shape and location (Fig. 4C). The A comparison of decomposition results from concurrently re-
shapes of the action potentials usually differ across motor units corded intramuscular (assumed gold standard) and surface EMG

Fig. 4. (A) One channel of surface EMG recorded from the biceps brachii muscle of a healthy man during a linear increase in force from 0% to 10% of the maximal voluntary
contraction (MVC) force in 10 s, followed by holding the 10% MVC force for 10 s. The surface EMG was recorded with a grid of 5  13 electrodes (8 mm interelectrode
distance) and was decomposed with the method proposed by Holobar and Zazula (2007). (B) Discharge times of 11 motor units that were identified from the decomposition
of the surface EMG. (C) The multi-channel action potentials for six representative motor units out of the 11 identified, as averaged over all the discharge times of each unit.
The amplitude values (RMS) for each channel (5  13) are also reported on the right of each action potential. The action potentials for these motor units all differed from each
other based on shape and location. MU, motor unit.
 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623 1621

Fig. 5. Surface and intramuscular EMG signals were concurrently recorded from the abductor digiti minimi muscle of a healthy man during an isometric contraction at 5% of
the maximal voluntary contraction (MVC) force. Surface EMG was detected with a grid of 5  13 electrodes (3 mm interelectrode distance) and intramuscular EMG by two
pairs of wire electrodes. The recordings were decomposed with the methods described by McGill et al. (2005) (intramuscular EMG) and by Holobar and Zazula (2007) (surface
EMG). Four motor units were detected by both the surface (sEMG) and the intramuscular EMG (iEMG) and the estimated discharge times were compared for the two
decomposition methods. The red tic marks indicate the discharge times identified by one of the two EMG techniques but not by the other (disagreement in the
decomposition). The sensitivity of surface EMG decomposition (reported on the right) was computed for each motor unit as the number of discharge times identified by both
surface and intramuscular EMG decomposition divided by the total number of discharges identified from the intramuscular EMG. The percent of false positives was computed
as the number of discharges identified by the surface but not by the intramuscular EMG decomposition divided by the total number of discharges identified from the
intramuscular EMG. The amplitude maps of the surface action potentials of the four motor units are shown above the discharge-time traces. MU, motor unit.

signals indicated a similar accuracy with the decomposition for the muscles, and conditions. For example, the range of individual mo-
two recordings during low-force isometric contractions (Fig. 5). For tor units detected from the abductor digiti minimi, tibialis anterior,
example, the proportion of discharge times identified in common and biceps brachii muscles during static contractions at 610% MVC
by the decomposition of concurrently recorded intramuscular force was 1–19 (Holobar et al., 2010). Simulation analyses indicate
and surface EMG signals was >90% for contraction forces ranging that the proportion of motor units identified by surface EMG
between 2.5% and 20% MVC force in the abductor digiti minimi, tib- decomposition is usually small when compared with the number
ialis anterior, and biceps brachii muscles (Holobar et al., 2010). of active motor units. Fig. 6 shows the motor units identified by
Moreover, it has been shown that the muscle architecture does the decomposition of a simulated surface EMG signal. In this exam-
not impact the quality of the EMG decomposition (Merletti et al., ple, approximately 23% of the active units were identified and
2008) and that high-density EMG decomposition methods have smaller percentages were obtained for higher contraction forces.
potential clinical applications (e.g., Kleine et al., 2008). Sample size is usually smaller with invasive methods, which can
impede generalizations about motor unit function (Tracy et al.,
2005). Despite the relatively few motor units that can be identified
4. Representativeness and accuracy
by decomposition of the surface EMG, the electrical activity of the
identified motor units explains most of the interference signal
The number of motor units that can be identified with decom-
(Fig. 6B). This finding indicates that the amplitude of the surface
position of the surface EMG varies substantially across subjects,

Fig. 6. (A) Simulated territories of active motor units in a muscle. Surface EMG signals generated by motor units were simulated with the planar volume conductor model
described by Farina and Merletti (2001) and the motor unit recruitment model described by Fuglevand et al. (1993). The model parameters were the same as in Keenan et al.
(2005) and the simulated signal-to-noise ratio was 20 dB. The EMG signals were generated during a linear increase and then decrease in force from 0% to 10% of the maximal
force and were detected with a grid of 5  13 electrodes with 3.5 mm interelectrode distance in both directions (bipolar derivations). The territories of the motor units
identified by the decomposition method are indicated as filled circles in (A) and constitute 23% of the active population. (B) The sum of the identified motor unit action
potential trains (grey line) compared with the signal derived from all the active motor units (black line). The simulated noise level corresponds to 20 dB of signal-to-noise
ratio. MUAP, motor unit action potential. Redrawn with permission from Holobar et al. (2009).
1622 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623

EMG does not represent the activity of a substantially larger num- lated with the force (Fig. 1C) and thus representative of the entire
ber of motor units than it is possible to identify through decompo- active population. Fig. 7 shows a representative analysis of the
sition. Furthermore, both amplitude and EMG decomposition are dimensionality of the multivariate smoothed discharge rates of
biased towards the motor units with the largest surface action 10 motor units in the abductor digiti minimi muscle during an iso-
potentials, i.e., large and superficial motor units. metric contraction at 10% of the maximal force. In this example,
Although both surface EMG amplitude and EMG decomposition one signal represented most of the variance (>70%) in the set of
provide a limited representation of muscle activity and EMG low-pass filtered discharge rates.
amplitude is a less accurate identification of the neural activation, Despite the possibility of inferring changes in the neural activa-
the decomposition of surface EMG is currently limited to low con- tion (mainly low-frequency components) from relatively few mo-
traction forces and to isometric conditions. These limitations do tor units in some conditions (Fig. 7), more subtle adjustments in
not apply to the analysis of neural activation with surface EMG motor neuron properties cannot be described as common changes
amplitude that can be easily performed in any conditions. in a population. For example, the changes in recruitment thresh-
Due to the relatively few motor units that are identified with olds and discharge rate during intermittent fatiguing contractions
decomposition of the surface EMG, the number of motor neuron differ substantially among low-threshold motor units (Farina
discharges is less than the output from the spinal cord. Moreover, et al., 2009). These changes cannot be inferred from only a limited
it is not possible to estimate the deficit in the estimation of number number of units and require a larger sample. Moreover, the
of discharges as this depends on factors that cannot be measured strength of correlation in low-frequency components of motor
or predicted. Consequently, it is not currently possible to obtain neuron discharges may vary across muscles (De Luca and Erim,
an absolute measure of the neural drive to the muscle by decompo- 1994). Currently, there are no validated methods that can identify
sition of the surface (or intramuscular) EMG, even at low contrac- most of the active motor units in vivo during a voluntary
tion forces. Nonetheless, relative changes in neural drive can be contraction.
estimated from the limited sample of motor units if the behavior
of the detected units is representative of the population; that is,
5. Conclusion
if the information contained in the ensemble of discharge patterns
has low dimensionality (Negro et al., 2009).
Surface EMG has been classically used to estimate the neural
Because the descending and sensory inputs diverge onto alpha
activation sent from the spinal cord to muscle by associating its
motor neurons (Lawrence et al., 1985; Ishizuka et al., 1979), there
amplitude to the number of action potentials discharged by a pop-
is some correlation between the low-frequency oscillations in mo-
ulation of motor neurons. Although this approach is valuable in
tor neuron discharge rates (De Luca et al., 1982). The low-fre-
some applications, the amplitude of the surface EMG is a relatively
quency oscillations in the motor output from the spinal cord can
crude index of neural drive and does not detect small fluctuations
often be described by a low-dimensional signal extracted from
in motor unit activity or adjustments during fatiguing contractions.
the correlated activity of relatively few motor neurons (Negro
One of the major advances in surface EMG processing in recent
et al., 2009). For this reason, the number of action potentials dis-
years is the development of techniques to identify the discharge
charged per second by relatively few units in Fig. 1 was well corre-
times of individual motor units from the interference signal, with
an accuracy that is similar to invasive methods. This approach,
however, is limited to isometric contractions and to low forces,
in contrast to the more widespread applications of surface EMG
recordings. An additional limitation of EMG decomposition is that
it identifies only a small proportion of the active units and these
tend to be located superficially in the muscle. Despite these limita-
tions, surface EMG decomposition allows the accurate detection of
relative changes in neural activation. As it seems unlikely that sur-
face EMG methods can be used to identify deep motor units, a
complete decoding of the neural drive to muscle in vivo will likely
require joint multi-channel intramuscular and surface EMG
recordings (Farina et al., 2008c; Holobar et al., 2010).

Acknowledgements

The authors are grateful to Francesco Negro and Jakob Lund

Dideriksen at Aalborg University for the help in the preparation
of Figs. 1, 2 and 7.
Grants
Partly supported by the European Project TREMOR (Contract #
224051) (DF), a Marie Curie reintegration grant within the 7th
European Community Framework Programme (iMOVE, Contract
Fig. 7. (A) Discharge times for 10 motor units identified in surface and intramus- No. 239216) (AH), Compagnia di San Paolo and Fondazione CRT
cular EMG recordings from the abductor digiti minimi muscle during a contraction (RM), and the National Institute on Aging (AG009000; RME).
at 10% of the maximal force of 10 s. (B) Smoothed discharge rates were obtained for
the 10 motor units by filtering the point process representing the inverse of the
interspike intervals (Hann window of 400 ms duration and high-pass filtering at References
0.75 Hz to remove the mean value). (C) Principal components extracted from the
smoothed discharge rates shown in (B). The procedure for this analysis is described Adrian E, Bronk D. The discharge of impulses in motor nerve fibres. II. The frequency
in Negro et al. (2009). The percent of variance of the smoothed discharge rates of discharges in reflex and voluntary contractions. J Physiol 1929;204:231–57.
explained by each principal component is reported on the right. The first Bigland-Ritchie B, Donovan EF, Roussos CS. Conduction velocity and EMG power
component explained most of the variance. MU, motor unit; PC, principal spectrum changes in fatigue of sustained maximal efforts. J Appl Physiol
component. 1981;51:1300–5.
 D. Farina et al. / Clinical Neurophysiology 121 (2010) 1616–1623 1623

Blok JH, van Dijk JP, Drost G, Zwarts MJ, Stegeman DF. A high-density multichannel Fuglevand AJ, Winter DA, Patla AE. Models of recruitment and rate coding
surface electromyography system for the characterization of single motor units. organization in motor-unit pools. J Neurophysiol 1993;70:2470–88.
Rev Sci Instrum 2002;73:1887–97. Garland SJ, Enoka RM, Serrano LP, Robinson GA. Behavior of motor units in human
Clancy EA, Hogan N. Single site electromyograph amplitude estimation. IEEE Trans biceps brachii during a submaximal fatiguing contraction. J Appl Physiol
Biomed Eng 1994;41:159–67. 1994;76:2411–9.
Clancy EA, Hogan N. Probability density of the surface electromyogram and its Hanson J, Persson A. Changes in the action potential and contraction of isolated frog
relation to amplitude detectors. IEEE Trans Biomed Eng 1999;46:730–9. muscle after repetitive stimulation. Acta Physiol Scand 1971;81:340–8.
Clancy EA, Morin EL, Merletti R. Sampling, noise-reduction and amplitude Holobar A, Zazula D. Multichannel blind source separation using convolution kernel
estimation issues in surface electromyography. J Electromyogr Kinesiol compensation. IEEE Trans Signal Process 2007;55:4487–96.
2002;12:1–16. Holobar A, Farina D, Gazzoni M, Merletti R, Zazula D. Estimating motor unit
Day SJ, Hulliger M. Experimental simulation of cat electromyogram: evidence for discharge patterns from high-density surface electromyogram. Clin
algebraic summation of motor-unit action-potential trains. J Neurophysiol Neurophysiol 2009;120:551–62.
2001;86:2144–58. Holobar A, Minetto MA, Botter A, Negro F, Farina D. Experimental analysis of
De Luca CJ, LeFever RS, McCue MP, Xenakis AP. Control scheme governing accuracy in the identification of motor unit spike trains from high-density
concurrently active human motor units during voluntary contractions. J surface EMG. IEEE Trans Neural Syst Rehabil Eng 2010. doi:10.1109/
Physiol 1982;329:129–42. TNSRE.2010.2041593.
De Luca CJ, Erim Z. Common drive of motor units in regulation of muscle force. Ishizuka N, Mannen H, Hongo T, Sasaki S. Trajectory of group is afferent fibers
Trends Neurosci 1994;17:299–305. stained with horseradish peroxidase in the lumbosacral spinal cord of the cat:
De Luca CJ, Adam A, Wotiz R, Gilmore LD, Nawab SH. Decomposition of surface EMG Three dimensional reconstructions from serial sections. J Comp Neurol
signals. J Neurophysiol 2006;96:1646–57. 1979;186:189–211.
Deschenes MR, Giles JA, McCoy RW, Volek JS, Gomez AL, Kraemer WJ. Neural factors Kallenberg LA, Hermens HJ. Behaviour of motor unit action potential rate, estimated
account for strength decrements observed after short-term muscle unloading. from surface EMG, as a measure of muscle activation level. J Neuroeng Rehabil
Am J Physiol Regul Integr Comp Physiol 2002;282:R578–83. 2006;3:15.
Dideriksen JL, Bækgaard M, Farina D, Enoka RM. A model of the surface EMG Keenan KG, Farina D, Maluf KS, Merletti R, Enoka RM. Influence of amplitude
amplitude at task failure. Soc Neurosci; 2009 [abstract]. cancellation on the simulated surface electromyogram. J Appl Physiol
Dimitrov GV, Disselhorst-Klug C, Dimitrova NA, Trachterna A, Rau G. The presence 2005;98:120–31.
of unknown layer of skin and fat is an obstacle to a correct estimation of the Keenan KG, Farina D, Merletti R, Enoka RM. Influence of motor unit properties on
motor unit size from surface detected potentials. Electromyogr Clin the size of the simulated evoked surface EMG potential. Exp Brain Res
Neurophysiol 2002;42:231–41. 2006;169:37–49.
Dimitrov GV, Arabadzhiev TI, Hogrel JY, Dimitrova NA. Simulation analysis of Kleine BU, Stegeman DF, Drost G, Zwarts MJ. Interspike interval analysis in a patient
interference EMG during fatiguing voluntary contractions. Part II – changes in with peripheral nerve hyperexcitability and potassium channel antibodies.
amplitude and spectral characteristics. J Electromyogr Kinesiol 2008;18:35–43. Muscle Nerve 2008;37:269–74.
Dimitrova NA, Dimitrov GV. Amplitude-related characteristics of motor unit and M- Lawrence DG, Porter R, Redman SJ. Corticomotoneuronal synapses in the monkey:
wave potentials during fatigue. A simulation study using literature data on light microscopic localization upon motoneurons of intrinsic muscles of the
intracellular potential changes found in vitro. J Electromyogr Kinesiol hand. J Comp Neurol 1985;232:499–510.
2002;12:339–49. McGill KC, Lateva ZC, Marateb HR. EMGLAB: an interactive EMG decomposition
Dimitrova NA, Dimitrov GV. Interpretation of EMG changes with fatigue: facts, program. J Neurosci Methods 2005;149:121–33.
pitfalls, and fallacies. J Electromyogr Kinesiol 2003;13:13–36. Merletti R, Knaflitz M, De Luca CJ. Myoelectric manifestations of fatigue in voluntary
Farina D, Merletti R. A novel approach for precise simulation of the EMG signals and electrically elicited contractions. J Appl Physiol 1990;69:1810–20.
detected by surface electrodes. IEEE Trans Biomed Eng 2001;48:637–46. Merletti R, Holobar A, Farina D. Analysis of motor units with high-density surface
Farina D, Cescon C, Merletti R. Influence of anatomical, physical, and detection- electromyography. J Electromyogr Kinesiol 2008;18:879–90.
system parameters on surface EMG. Biol Cybern 2002;86:445–56. Mottram CJ, Jakobi JM, Semmler JG, Enoka RM. Motor-unit activity differs with load
Farina D, Merletti R, Enoka RM. The extraction of neural strategies from the surface type during a fatiguing contraction. J Neurophysiol 2005;93:1381–92.
EMG. J Appl Physiol 2004a;96:1486–95. Negro F, Holobar A, Farina D. Fluctuations in isometric muscle force can be
Farina D, Mesin L, Martina S, Merletti R. A surface EMG generation model with described by one linear projection of low-frequency components of motor
multilayer cylindrical description of the volume conductor. IEEE Trans Biomed neuron discharge rates. J Physiol 2009;587:5925–38.
Eng 2004b;51:415–26. Person RS, Kudina LP. Discharge frequency and discharge pattern of human motor
Farina D, Cescon C, Negro F, Enoka RM. Amplitude cancellation of motor-unit action units during voluntary contraction of muscle. Electroencephalogr Clin
potentials in the surface electromyogram can be estimated with spike-triggered Neurophysiol 1972;32:471–83.
averaging. J Neurophysiol 2008a;100:431–40. Riley ZA, Maerz AH, Litsey JC, Enoka RM. Motor unit recruitment in human biceps
Farina D, Negro F, Gazzoni M, Enoka RM. Detecting the unique representation of brachii during sustained voluntary contractions. J Physiol 2008;586:2183–93.
motor-unit action potentials in the surface electromyogram. J Neurophysiol Tracy BL, Maluf KS, Stephenson JL, Hunter SK, Enoka RM. Variability of motor unit
2008b;100:1223–33. discharge and force fluctuations across a range of muscle forces in older adults.
Farina D, Yoshida K, Stieglitz T, Koch KP. Multichannel thin-film electrode for Muscle Nerve 2005;32:533–40.
intramuscular electromyographic recordings. J Appl Physiol 2008c;104:821–7. Yang JF, Winter DA. Electromyographic amplitude normalization methods:
Farina D, Holobar A, Gazzoni M, Zazula D, Merletti R, Enoka RM. Adjustments differ improving their sensitivity as diagnostic tools in gait analysis. Arch Phys Med
among low-threshold motor units during intermittent, isometric contractions. J Rehabil 1984;65:517–21.
Neurophysiol 2009;101:350–9.