A coherence study on EEG and EMG signals

Giulia Cisotto, Umberto Michieli, Leonardo Badia

Dept. of Information Engineering, University of Padova
via Gradenigo 6B, 35131 Padova, Italy
email: {giulia.cisotto, leonardo.badia}@dei.unipd.it

Abstract—The aim of this study is to investigate bursts- central, i.e. conscious, control has been suggested [3] [4] [5],
related EEG signals in a focal hand dystonia patient. Despite
arXiv:1712.01277v1 [physics.med-ph] 4 Dec 2017

but not consistently proved and accepted, yet.

of considering time domain and frequency domain techniques as
mutually exclusive analysis, in this contribution we have taken The Jerk-locked back averaging (JLBA) technique has been
advantage from both of them: particularly, in the frequency effectively employed on MRCP to identify the central origin
domain, coherence was used to identify the most likely frequency of a specific kind of involuntary movements observed in
bands of interaction between brain and muscles; then, in the time myoclonus, Tourette’s syndrome and other psychogenic motor
domain, cross-correlation was exploited to verify the physiological disorders [6]. Thanks to such technique, a sharp biphasic wave-
reliability of such a relationship in terms of signal transmission
delay from the centre to the periphery. Our preliminary results form could be consistently seen on the averaged EEG signal,
suggest - in line with recent literature - that activity in the high β especially at the central and contralateral areas. Moreover, this
band (around 30 Hz) could represent an electroencephalographic EEG potential anticipated the EMG onset by 15 to 20 ms on
correlate for the pathological electromyographic bursts affecting average [7] [8].
the focal hand dystonia condition. Even though a future study
on a larger sample is needed to statistically support these
Later on, other techniques have been utilized to quantify
preliminary findings, this contribution allows to think of new the influence of brain activity on the motor output. Moreover,
kinds of rehabilitation from focal hand dystonia that could target coherence between EEG and EMG has been computed in
the actual electroencephalographic correlate of the pathology, i.e. many studies for different kinds of patients: widely-known as
phenotypically expressed by bursts, with the consequence of a cortico-muscular coherence (CMC) [9], it is usually evaluated
relevant functional improvement.
during sustained contractions at a predetermined level, e.g. 15
to 20 percent of the maximal voluntary contraction, in order
I. I NTRODUCTION to ensure stable motor units engagement.
There is a wide literature robustly reporting on how a However, CMC was also employed in few recent works on
voluntary motor output is prepared and driven by the central Parkinson’s disease resting tremor [10][11]: in those studies,
nervous system, the sensorimotor circuit, in particular. Brain no movement was accomplished by the patients but their pos-
signals recorded at several depths of the brain, i.e. (from tural tremor was recorded by EMG together with synchronous
outer to inner layer) the scalp, the dura, the cortex and the EEG or magnetoencephalographic (MEG) activity.
basal ganglia, have shown robust patterns of activation and A clear peak of activity at the frequency of tremor and its
deactivation in specific regions, frequency bands and time second harmonic, around 5 Hz and 10 Hz respectively, could
periods. be seen in the EMG power spectrum; moreover, CMC showed
Specifically, power decrease at the contralateral hand-related a statistically significant peak of coherence between the EEG
side (to the movement) in the so-called µ and β bands, around signal recorded from the contralateral hand-related scalp area
10 Hz and 20 Hz, respectively, occurs as soon as 1s before and the fingers extensor muscle.
movement onset. This phenomenon is known as event-related Therefore, in case of rhythmic pathological behaviour, a
desynchronization (ERD) [1]. relationship between central and peripheral activity could be
In the time domain, time-locking each electroencephalo- significantly quantified at rest, too.
graphic (EEG) signal on its corresponding electromyographic The aim of this study is to apply a similar concept to the
(EMG) activation onset, and averaging among several re- investigation of bursts-related EEG signals in one focal hand
sponses, a typical waveform could be observed: indeed, the dystonia (FHD) patient.
so-called readiness potential starts as soon as 1.5 to 1 s before FHD is a movement disorder that causes people who are
movement onset with a slow decrease of signal amplitude; affected by it to experience an abnormal and involuntary
other known components follow, each of them with a specific co-contraction of the agonist and antagonist muscles of the
clinical meaning. This complex behavior, overall, is labelled hand and the forearm. It has been shown to originate in the
as movement-related cortical potential (MRCP) [2]. central nervous system and to cause abnormal patterns of brain
In case of neuro-motor pathologies, ERD and MRCP could activation, especially in the β band (as suggested by recent
become carriers of important information related to abnormal literature) [12].
behaviours of the patient. Particularly, in case of motor dis- Bursts are abrupt and giant involuntary muscular contraction
orders, where movements are often involuntarily produced, a events that typically affect EMG of this kind of patients,
especially at rest when they largely exceed the very low A. Frequency domain analysis: cortico-muscular coherence
background activity. The coherence of two discrete-time signals x[m] and y[m],
In this contribution, CMC as well as cross-correlation have regarded as stochastic processes, is given by:
been computed between EEG and EMG to assess, both in Pxy (f )
the frequency and in the time domain, the effect of a central Cohxy (f ) , p p , (1)
driver onto the motor output. Our preliminary results show |Px (f )| · |Py (f )|
the influence of EEG on pathological EMG oscillatory activ- where Px (f ) is the PSD of x[m] and Pxy (f ) =
n
ity. Coherence was employed to identify the most involved 1X
frequencies, while cross-correlation was used to support the Xi (f )Yi∗ (f ) is the cross-power spectral density (CPSD)
n i=1
physiological meaning of such EEG-EMG relationship. between x[m] and y[m].
In the rest of the paper, section II will present the methods of In order to provide a statistically significant result, a confi-
the study, section III the most interesting preliminary results, dence level CL of 95 %, i.e. a critical level of α = 0.05, was
while the final section IV will discuss them in comparison obtained from the following formula [13]:
with existing literature on the topic along with an overview of 1

some limitations to be overcome in the future; a perspective CL = 1 − (1 − α) N −1 , (2)

view for clinical applications in motor rehabilitation will be where N is the number of signal segments used to estimate
provided, also. the coherence value.
The PSD, the CPSD and the coherence values were esti-
mated via the Fast Fourier Transform (FFT)-based Welch’s
II. M ATERIALS AND M ETHODS method: specifically, the signal length was set to L = 1024
samples (1.024 s) and the number of FFT points to 1024
One FHD patient and one healthy subject (HS) were in- samples. Border effects were mitigated by a Hann sliding
volved in this pilot study. The EEG was recorded from one windowing with overlap of 50 % [14].
monopolar EEG channel placed on C3, the standard location
B. Time domain analysis: cross-correlation function
of the International 10-20 System over the left hemisphere
corresponding to the brain region related to the functioning Generally speaking, given two discrete-time signals x[m]
of the (dominant) right-hand. The EMG was recorded from and y[m], their cross-correlation function is defined as:
one bipolar channel placed on the abductor pollicis brevis, +∞
X
the intrinsic hand muscle responsible for the abduction of the rxy [n] , x∗ [m]y[n + m]. (3)
thumb. Both signals were sampled with a sampling frequency m=−∞
of 1 kHz and quantized at 16 bit. In the experiment, the Cross-correlation is particularly useful to evaluate the simi-
participants were sitting quietly on a comfortable chair in front larity between two signals as a function of the time shift n
of a screen placed 1 meter apart from them, on a table. They (expressed in number of samples) of the second signal behind
were simply required to rest with opened eyes for about 3 the first one.
minutes with their limbs laying on the table in front of them. In the present analysis, the absolute value of the correlation
At a first glance, it was possible to assess a clear difference between the EEG and the EMG signals computed at its
between the two EMG signals: in the FHD patient, the maximum and normalized by the square root of the product
amplitude of the signal assumed values up to ± 200 µV, while of the signals energies Ex and Ey was evaluated. Therefore,
its power spectral density (PSD) took significant values in the quantity:
the frequency band [5, 200] Hz. However, in case of HS, the max(rxy [n])
rmax = p . (4)
amplitude of the EMG signal did not exceed ± 20 µV, with a Ex Ey
significant PSD extended from 5 to 50 Hz. was considered as a measure of similarity between the two
In the offline analysis, signals were preprocessed to limit signals.
their frequency range in the frequency band of interest. Specif- Moreover, the lag n which the maximum was found at was
ically, the EEG was filtered through an elliptic filter of order taken into account as a measure of the transmission delay
24 with a passband of [4, 45] Hz. The EMG was processed by from the brain to the muscle, i.e. the time taken for a motor
a high-pass elliptic filter of order 11 with cut-off frequency command to travel from its origin in the central nervous system
at 5 Hz. A series of notch filters of order 14 were used with to the target effector at the periphery.
cut-off frequencies at 50 Hz and subsequent harmonics up to Particularly, 71 pairs of EEG and EMG signals were ex-
350 Hz were put to reduce the effect of the mains. tracted from the whole EEG and EMG recordings of the FHD
Then, CMC as well as cross-correlation have been computed patient. They were selected empirically as examples of bursty
between the EEG (otherwise labelled as signal x[m]) and the EMG activity (with their corresponding EEG). The duration
EMG (otherwise labelled as signal y[m]) signals, in order to of these signals was variable (0.70 ± 0.66 s): all of them were
assess the quantitative relationship between them, both in the included in the analysis to keep into account the variability of
frequency domain and in the time domain. the burst events.
 0.3
Coherence Coherence
CL 1 CL

0.25
0.8

Amplitude
0.2
0.6
Amplitude

0.15 0.4

0.2
0.1

0
0 5 10 15 20 25 30 35 40 45 50
0.05 Frequency [Hz]

Fig. 3. CMC, in absolute value, between chunks of healthy-like EEG and

0
0 5 10 15 20 25 30 35 40 45 50 EMG for the pathological subject (CL = 0.776 with N = 3 and α = 0.05).
Frequency [Hz]

0.8
Fig. 1. Absolute value of CMC for the healthy participant (CL = 0.067 Coherence
CL
with N = 44 and α = 0.05). 0.7

0.06 0.6
Coherence
CL
0.5

Amplitude
0.05
0.4

0.04 0.3
Amplitude

0.2
0.03
0.1

0.02 0
0 5 10 15 20 25 30 35 40 45 50
Frequency [Hz]

0.01
Fig. 4. CMC, in absolute value, between chunks of bursts-affected EEG and
EMG for the pathological subject (CL = 0.451 with N = 6 and α = 0.05).
0
0 5 10 15 20 25 30 35 40 45 50
Frequency [Hz]

Fig. 2. Absolute value of CMC for the pathological subject (CL = 0.008 probably due to the larger bandwidth of the pathological EMG
with N = 393 and α = 0.05). of the patient, as mentioned in the previous section.
In order to confirm our hypothesis, we selected a portion
of the whole recorded data where bursts mostly affected
To support the physiological meaning of the EEG-EMG the EMG signal and evaluated the CMC in this particular
coherent components, the cross-correlation function was com- case. As a further support, we selected another portion of
puted between the narrow-band EEG signals filtered in the EMG signal where healthy-like activity could be observed
high β band, i.e. between 26 and 31 Hz, and the EMG signal and computed CMC as well. Two typical examples of both
limited to 250 Hz by a band-pass filter with frequency band situations are reported next. Fig. 3 shows the coherence result
[5, 250] Hz. when comparing two chunks of the EEG and EMG signals for
the healthy-like case. Here, two main peaks can be seen at the
III. R ESULTS
frequencies of 8 Hz and 18 Hz, but no significant coherence
A. EEG-EMG coherence values at frequencies higher than 30 Hz.
In this section the results are shown in regards to the CMC On the contrary, Fig. 4 reports the coherence spectrum in
measure for both the HS and the FHD patient. In the case of case of bursts-affected chunks. Significantly, the figure shows
the healthy participant, the CMC spectrum could be seen in that coherence values at low frequencies are heavily reduced,
Fig. 1. It has to be noted that peaks above the confidence level whereas some peaks around 20 and 35 Hz appeared, hence
can be observed in the frequency range between 5 and 20 Hz, the hypothesis that higher frequencies components are related
only, with a particularly strong coherence at 20 Hz. to bursty EMG activity in the FHD patient could actually be
On the other hand, the CMC spectrum of the FHD patient supported.
is reported in Fig. 2. It can be observed that a larger frequency
band contribute to the coherence between EEG and EMG B. EEG-EMG cross-correlation
signals. It is also important to highlight the presence of peaks As mentioned above, cross-correlation of EEG and EMG
in the upper side of the spectrum, i.e. [20, 45] Hz. This is was computed to investigate the physiological reliability on
 Istogramma del modulo del massimo valore normalizzato della correlazione
12
20 X: -0.01165
Y: 19

18
10
16
number of occurrences

8 14

number of occurences
12

6 10

8
4
6

4
2

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -1.5 -1 -0.5 0 0.5 1 1.5 2
maximum value time [s]

Fig. 5. Empirical distribution of the maximum value of the cross-correlation Fig. 6. Empirical distribution of the lags where the maximum of the cross-
function of the 71 pairs of EEG and EMG signals of the FHD patient. correlation was found.

the relationship between the high β band EEG component some technical questions in its application to FHD, as reported
with the EMG. by other literature [15][16]. Specifically, the following critical
Fig. 5 reports the empirical distribution of the maximum points were highlighted:
values of the cross-correlation function found from the 71 pairs • high-frequency and irregularity of the jerks might prevent
of EEG and EMG signals. Mean value was found of 0.683, their correct identification (EMG activation should be
with variance of 0.0293. This result certainly shows a strong absent for 100ms at least before the EMG burst);
connection between the narrow-band EEG and the EMG, as • similarly, the absence of giant somatosensory evoked
indicated by the high average value. components in many subjects could prevent success of
Finally, Fig. 6 displays the empirical distribution of the lag the JLBA procedure;
where the maximum value of the cross-correlation function • frequency domain-based analysis have been successfully
of the 71 pairs of EEG and EMG signals was found. Mean proposed to identify EEG correlates of myoclonic jerks,
value occurred at −11.65 ms. As neural impulses propagate even in case where time domain JLBA has failed.
at a speed of about 100 m/s, the transmission of signals from Since similarity could be noted between EMG bursts seen
the brain to the hand muscles could be estimated of about 10 in FHD and jerks present in myoclonus, many considerations
ms, which is in line with the results we achieved. explained above hold true for analysis of EMG bursts in case
The standard deviation is considerably high (it was found of FHD.
to be about 100 ms) but this could be explained because of Therefore, in this study we proposed the use of CMC in case
the limited size of the data sample. Indeed, we expect that the of FHD, too. Besides, we do not believe time and frequency-
tendency observed in this study could be further confirmed domain analysis to be mutually exclusive but, on the contrary,
(with a reduced standard deviation), with an increased sample we claim the opportunity to use them to complement each
size. other. Therefore, in this contribution, coherence analysis was
used to identify the most likely frequency bands of central-
IV. D ISCUSSION
periphery communications; then, cross-correlation function
FHD and other movement disorders show jerks during was employed to verify the physiological reliability of such
movement or even during rest. Identifying possible central relationship (as suggested by [7]).
drivers of such abnormal muscular activity is a very relevant In line with well-known literature about CMC [17] [13],
issue and could represent a key aspect for improving diagnosis we found a peak-component at 20 Hz (β band) together with
and rehabilitation. JLBA is the only technique that could lower peaks at lower frequencies (θ and α bands) in the HS.
directly reveal a causal relationship between an average brain On the contrary, the FHD patient showed different CMC
activity and an average muscular response in many patholog- patterns in different behavioral scenarios: specifically, when
ical cases. a healthy-like recording period is selected, CMC spectrum
However, JLBA is not always applicable and requires many displayed two major peaks, one around 5 Hz and the other
chunks of EEG and EMG to be robust. Moreover, identification one at 20 Hz. In this case, we could advance the hypothesis -
of jerks and averaging among jerk-locked EMG signals raise supported by literature [10] [11] - that the slower component
is related to the postural tremor affecting this subject, while R EFERENCES
the second one seems to reflect the healthy-like β band [1] G. Pfurtscheller and F. H. Lopes Da Silva, “Event-related EEG/MEG
synchronous activity between EEG and EMG that has been synchronization and desynchronization: Basic principles,” Clinical Neu-
seen in the control subject, as well. Worthy of notice is the fact rophysiology, vol. 110, no. 11, pp. 1842–1857, 1999.
[2] H. Shibasaki, G. Barrett, E. Halliday, and A. Halliday, “Components
that the tremor-related component significantly prevails onto of the movement-related cortical potential and their scalp topography,”
the other one and it is the only one to exceed the confident Electroencephalography and Clinical Neurophysiology, vol. 49, no. 3,
level of 95 %. When an EMG period heavily characterized by pp. 213–226, 1980.
[3] P. T. Lin and M. Hallett, “The pathophysiology of focal hand dystonia,”
the presence of bursts is selected, we could observe two major Journal of Hand Therapy, vol. 22, no. 2, pp. 109–114, 2009.
frequency components in the CMC spectrum: the first one was [4] M. H. Ruiz, P. Senghaas, M. Grossbach, H.-C. Jabusch, M. Bangert,
around 20 Hz (β band), it exceeded the confidence level of 95 F. Hummel, C. Gerloff, and E. Altenmuller, “Defective inhibition and
inter-regional phase synchronization in pianists with musician’s dysto-
% and it could resemble the healthy-like β band component nia: an EEG study.,” Human brain mapping, vol. 30, pp. 2689–700, Aug
of the normal condition. The second one, interestingly, was 2009.
significant at the 95 % as the first component, and could [5] H. Shibasaki and M. Hallett, “Electrophysiological studies of my-
oclonus,” Muscle and Nerve, vol. 31, no. 2, pp. 157–174, 2005.
be suggested to represent the correlate of the bursty activity [6] H. Shibasaki and Y. Kuroiwa, “Electroencephalographic correlates of
observed in the pathological EMG of the patient. myoclonus,” Electroencephalography and Clinical Neurophysiology,
An average delay of 10 ms of EMG behind EEG has been vol. 39, no. 5, pp. 455–463, 1975.
[7] H. Shibasaki, “Cortical activities associated with voluntary movements
observed through the cross-correlation analysis and it is in line and involuntary movements,” Clinical Neurophysiology, vol. 123, no. 2,
with physiology and other literature [13]. pp. 229–243, 2012.
Overall, our preliminary results claimed that activity in the [8] G. Avanzini, H. Shibasaki, G. Rubboli, L. Canafoglia, F. Panzica,
S. Franceschetti, and M. Hallett, “Neurophysiology of myoclonus
high β band [25,35] Hz could represent an EEG correlate for and progressive myoclonus epilepsies,” Epileptic Disorders, vol. 18,
the EMG bursts affecting the FHD condition. no. September, pp. 11–27, 2016.
Even though the small size of the data sample could [9] T. Mima, S. Ohara, and T. Nagamine, “Corticalmuscular coherence,”
International Congress Series, vol. 1226, pp. 109–119, 2002.
be a limitation to the study, our conclusions were primarly [10] L. Timmermann, J. Gross, M. Dirks, J. Volkmann, H. J. Freund, and
driven by our data analysis, but their reliability was strongly A. Schnitzler, “The cerebral oscillatory network of parkinsonian resting
supported by existing literature about similar studies accom- tremor,” Brain, vol. 126, no. 1, pp. 199–212, 2003.
[11] B. Hellwig, S. Häußler, B. Schelter, M. Lauk, B. Guschlbauer, J. Tim-
plished with larger samples of data and patients, even though mer, and C. Lücking, “Tremor-correlated cortical activity in essential
in slightly different diseases, e.g.myoclonus and Parkinson’s tremor,” The Lancet, vol. 357, no. 9255, pp. 519–523, 2001.
disease. [12] F. Tecchio, J. M. Melgari, F. Zappasodi, C. Porcaro, D. Milazzo,
E. Cassetta, and P. M. Rossini, “Sensorimotor integration in focal
Even though a future study on a larger sample is needed to task-specific hand dystonia: A magnetoencephalographic assessment,”
statistically support these preliminary findings, this contribu- Neuroscience, vol. 154, no. 2, pp. 563–571, 2008.
tion allows to think of new kinds of rehabilitation interventions [13] T. Mima and M. Hallett, “Electroencephalographic analysis of cortico-
muscular coherence: reference effect, volume conduction and generator
[18] [19] for focal hand dystonia patients that could target mechanism,” Clinical Neurophysiology, vol. 110, no. 11, pp. 1892–1899,
the actual EEG correlate of the pathology, i.e. phenotypically 1999.
expressed by bursts, with the consequence of a relevant [14] S. K. Mitra, Digital signal processing: a computer-based approach.
McGraw-Hill, 2007.
functional improvement. [15] P. Brown, S. F. Farmer, D. M. Halliday, J. Marsden, and J. R. Rosenberg,
“Coherent cortical and muscle discharge in cortical myoclonus,” Brain,
V. C ONCLUSIONS vol. 122, no. 3, pp. 461–472, 1999.
[16] Y. Xu, M. V.M., C. Z., and M. K.R., “Cortico-Muscular Coherence with
The aim of this study is to investigate bursts-related EEG Time Lag with Application to Delay Estimation.,” IEEE Transactions
signals in a focal hand dystonia patient. Despite of considering on Biomedical Engineering, May 2016.
time domain and frequency domain techniques as mutually [17] T. Mima and M. Hallett, “Corticomuscular coherence: a review.,”
Journal of Clinical Neurophysiology, vol. 16, no. 6, pp. 501–511, 1999.
exclusive analysis, in this contribution we have taken advan- [18] K. von Carlowitz-Ghori, Z. Bayraktaroglu, G. Waterstraat, G. Curio,
tage from both of them: particularly, in the frequency domain, and V. V. Nikulin, “Voluntary control of corticomuscular coherence
cortico-muscular coherence was used to identify the most through neurofeedback: A proof-of-principle study in healthy subjects,”
Neuroscience, vol. 290, pp. 243–254, 2015.
likely frequency bands of interaction between brain and mus- [19] Y. Hashimoto, T. Ota, M. Mukaino, M. Liu, and J. Ushiba, “Functional
cles; then, in the time domain, cross-correlation was exploited recovery from chronic writer’s cramp by brain-computer interface reha-
to verify the physiological reliability of such a relationship bilitation: a case report.,” BMC neuroscience, vol. 15, p. 103, jan 2014.
in terms of signal transmission delay from the centre to the
periphery. The most interesting result suggested that the high
β band activity in the EEG could be responsible for the
bursty activity observed in the EMG. Even though a future
study on a larger sample is needed to statistically support
these preliminary findings, this contribution allows to think
of new kinds of rehabilitation interventions for focal hand
dystonia patients that could target the actual EEG correlate
of the pathology with consequence improvement of the motor
functions.