Professional Documents
Culture Documents
A Wavelet Based Approach For The Detection of Coupling in EEG Signals
A Wavelet Based Approach For The Detection of Coupling in EEG Signals
1 2 4
Department of Electrical and Department of Medicine (Neurology) Sensory Motor Performance
3
Computer Engineering, Pacific Parkinson's Research Centre, Program (SMPP)
The University of British The University of British Columbia , Rehabilitation Institute of Chicago
Columbia, Vancouver, BC Vancouver, BC Chicago, IL
Canada V6T 1Z4 mmckeown@intechange.ubc.ca L-myers2@northwestern.edu
rayans@ece.ubc.ca
Abstract—Electroencephalograms (EEGs) provide a non- approach by building it a theoretical point of view and
invasive way of measuring brainwave activity from sensors providing a study of its statistical significance. The use of
placed on the scalp. In this paper we present an approach to Fourier coherence in neurophysiology is extensive [5][6]. The
measure coupling, or synchrony, between various parts of the
latter utilizes a short time Fourier transform (STFT) approach
brain, critical for motor and cognitive processing, using wavelet
coherence of EEG signals. We provide an argument, highlighting to coherence.
the benefits of using this approach as opposed to the regular In the rest of this paper, we briefly present coherence as a
Fourier based coherence, in the context of localizing short measure of coupling, and also introduce STFT extension to
significant bursts of coherence between non-stationary EEG coherence as one of the proposed solutions to the problem of
signals, to which regular coherence is insensitive. We further non-stationarity. We then introduce wavelet coherence, and
highlight the benefits of the wavelets approach by exploring how
present its formulation as a natural and logical extension to
a single time-frequency coherence map can be controlled to yield
various time and/or frequency resolutions. coherence measures, and explain its benefits over the standard
STFT approach. We present the results to verify our
arguments using real EEG data collected from a human
I. INTRODUCTION subject performing an isometric motor task. We compare
standard STFT results with the wavelet approach based on
Understanding of cerebral dynamics is aided by the ability
statistical significance.
to track temporal coupling between various neural signals [1].
The most common way of measuring coupling is through the
II. COHERENCE
use of the coherence metric. Coherence is the degree of time-
locked correlation between two signals as a function of A. Regular Coherence and its Statistical Significance
frequency [2]. The problem that we address in this paper, and
Coherence as a measure is an extension to Pearson’s
suggest a solution for, is that coherence assumes stationarity
correlation coefficient, and is defined as the absolute square of
of the signals in question, and does not show changes in
the cross-spectrum of two signals normalized by the product
coupling with time. We claim and verify that through the use
of their auto-spectra [6]. Thus one way of calculating
of a wavelets extension to classical Fourier coherence, one can
coherence between two signals x(t) and y(t) is as follows:
effectively measure the dynamic behavior of coupling, thus
First, calculate the Fourier transform of each of the signals
observing how changes in task, for example, affect coupling
over multiple contiguous segments (overlapping or
between various locales in the brain. We believe this tool has
contiguous).
widespread applicability for investigating the relationships
between its various cortices. (1)
∫
− j 2 π ft
X i ( f ) = 1 / 2π x (t )e dt
0-7803-8709-0/05/$20.00©2005 IEEE
v
signals modulated by a smooth window function H, to reduce approach are given in [1].
problems such as spectral leakage.
Next, calculate the auto-spectrum of each signal and their III. WAVELET COHERENCE
cross-spectrum over each segment.
A. Wavelets
Pxx ( f , i ) = X i ( f ) X i* ( f ) (3)
In contrast to the short time Fourier transform, which is an
Pxy ( f , i ) = X i ( f )Yi* ( f ) (4) extension of the regular Fourier transform, adapted to handle
non-stationarity, the wavelet transform is inherently a time-
Pyy ( f , i ) = Yi ( f )Yi * ( f ) (5) frequency analysis of a signal. It is similar to the STFT in that
In (3), (4) and (5) Pxx and Pyy represent the auto-spectra it uses finite duration basis functions to estimate the transform,
while Pxy represents the cross spectrum, with the * operator but different in that the length of the support is a function of
indicating complex conjugation. frequency. Thus while the STFT uses a fixed size rectangular
Finally, calculate the coherence between the two signals. tiling of the time-frequency plane, the wavelet transform
(Use the average of the spectra over the segments or utilizes rectangles of variable dimensions but constant area.
windows). Naturally, they are still governed by the uncertainty principle.
2
∑P i
xy ( f , i)
K( f ) = . (6)
∑Pi
xx ( f , i )∑ Pyy ( f , i )
i
Using the Schwarz inequality, observe that the coherence
takes on a value between 0 and 1. Also note here that the
coherence (K) can be considered 95% significant if it exceeds Fig.1.The time frequency tiling of the STFT (left) and the wavelet
the following threshold. [7] transform (right).
approach is generally used to generate auto- and cross- Where x(t) is the signal to be transformed, Ψ the wavelet, b
spectrogams, which are in turn utilized to produce a and s, are the time shift and scale respectively. Thus,Ψ((u-
‘coherogram’. The coherogram is coherence calculated around b)/s) is a shifted and scaled version of a mother wavelet Ψ(u).
a number of time instants. It results in a three dimensional In general, there exists an inverse relation between scale and
matrix of time and frequency versus coherence. frequency; frequency is inversely proportional to scale. The
The problem with such an approach is that, while it removes wavelet we have chosen for this work is the complex Morlet
the assumption of wide sense stationarity inherent in the wavelet [9].
coherence measure, it still requires stationarity within each
1 iπ f c u −u 2
/ fb
time interval for which coherence is calculated. More Ψ (u ) = e e (9)
importantly, in practice it requires careful design of several πfb
parameters, namely section length (over which each coherence
Where fc is the wavelet center frequency and fb is its
estimate is measured), window length and overlapping (within
bandwidth (we select these parameters to be 1 and 5
each coherence estimate), which affects the resolution of the
respectively). One can observe from (9) that the Morlet
coupling measure. Thus the STFT approach requires a priori
wavelet is a complex sinusoid multiplied by a Gaussian
information about the coupling range in time and frequency, in
window. The size of the window increases with frequency,
order to allocate the time-frequency resolution which is
while preserving the same number of oscillations across all
constrained by the uncertainty principle: the wider the
frequencies, which is how it differs from the STFT. The STFT
windows, the better the frequency resolution, at the expense of
preserves window sizes while varying the number of
timing information, and vice versa. More details on this
oscillations within each window.
vi
B. Wavelet Coherence x 10
-4
EEG Signal Recorded Over the Primary Motor Cortex
2
Following [1], we first define the wavelet cross spectrum:
* 1
PW xy (t , f ) = W x (t , f )W y (t , f ) . (10)
0
The definition of the auto-spectrum follows naturally from
this, as in the preceding section on coherence. Next we -1
∑ PW
-4
EEG Signal Recorded Over the Supplementary Motor Cortex
xy (T , f , i ) 4
x 10
i =− ∆
C (t , f ) = , (11) 2
∑ PW
i
xx (T , f , i )∑ PW yy (T , f , i )
i
0
IV. EXPERIMENT
100, corresponding to 0.5 seconds, and overlapping of half the
A. Methods window size. The results are displayed in fig.3. Fig.3 indicates
EEG was recorded from the scalp using 61 electrodes. Two that the strongest coupling is around 30Hz, however it gives
channels located over the primary motor cortex (PMC) and no information as to how that coupling is distributed over the
one channel located over the supplementary motor cortex 45-second duration of the experiment. Thus, to gain more
(SMC) were selected for coherence, STFT coherence and insight into the temporal distribution of the coherence, we
wavelet coherence measurements. The subject was resort to computing both the STFT coherogram and the
performing an isometric task. Data length was 45 seconds and Wavelet coherogram. Fig.4 displays the result of the STFT
onset of the task was at approximately 5 seconds. Data were coherogram, after we removed the insignificant peaks. We
sampled and digitized at 2000Hz. The entire EEG data set was evaluated significance based on (7), calculated at each time
subjected to a Laplacian Spatial Filter using commercial instant, and verified via Monte Carlo simulations on white
software[10]. It was then bandpass filtered between 0.5 and noise [1]. The STFT coherence was based on a segment size
70Hz and downsampled to 200Hz to reduce the computational of four seconds
load. Figure 2 displays the time domain signals after the Next, we perform the wavelet coherogram as in (11), to
preprocessing steps. obtain the result in fig.5 showing only the significant peaks.
Significance was tested based on the tabulated values reported
B. Results in [1]. Fig.5 shows the high temporal resolution exhibited by
We computed regular coherence between the first PMC the wavelet based coherence. It shows us that the coherence in
channel and the SMC channel as in (6) using a window size of the 30Hz band, previously shown to be related to ongoing
vii
wavelet Coherogram
muscle contraction is in fact not completely continuous. This 70
Frequency
The results of this smoothing operation can be seen in fig.6. 40
STFT Coherogram
70 10
0 5 10 15 20 25 30 35 40
Time
wavelet Coherogram
40 70
30 60
20 50
Frequency
10 40
0 5 10 15 20 25 30 35 40 45
Time
10
To further test the wavelet coherence method, in 0 5 10 15 20
Time
25 30 35 40
comparison with the STFT coherence method, we next Fig.6. Wavelet coherence displayed after smoothing over a span of four
evaluate both of them over another pair of channels. This time seconds. Compare with fig. 4.
we choose a second channel from the primary motor cortex
more anterior than the first, with the same supplementary 0.46
Average Wavelet Coherence
0.42
C. Discussion 0.4
Upon comparing fig.4 and fig.6, one notices that the wavelet 0.38
viii
some instances where 10Hz coherence exists. In conclusion, we have shown how the wavelets
wavelet Coherogram
coherence performs favorably compared to the STFT
70 coherence approach at detecting coupling at various
frequency ranges, especially when the potentially
60
‘interesting’ frequencies cannot be specified a prior. As
50 both methods reveal that the EEG-EEG coupling between
brain regions are transient, task-dependent, and take place at
40
multiple frequencies, the wavelet coherence may prove
Frequency
30
advantageous because it is more versatile in modifying the
temporal resolution to observe long-term trends in coupling.
20
REFERENCES
10
[1] J. P. Lachaux, et al., “Estimating the time-course of coherence between
0
single-trial brain signals: an introduction to wavelet coherence,”
0 5 10 15 20 25 30 35 40 Neurophysiol. Clin , 32(3),2002, pp.157-74.
Time
[2] J.N. Caviness, C.H. Adler, M.N. Sabbagh, D.J. Connor, J.L. Hernandez,
Fig.9. Wavelets coherence of the second primary motor cortex T.D. Lagerlund, “Abnormal corticomuscular coherence is associated
channel, with the supplementary motor cortex channel, plotted after with the small amplitude cortical myoclonus in Parkinson's disease,”
smoothing. The darker regions indicate higher levels of coherence. Mov Disord.,18(10), Oct 2003, pp. 1157-62.
[3] C. Torrence, and G.P. Compo, “A Practical Guide to Wavelet Anaysis,”
Bulletin of the American Meteorogical Society, vol.79 (1), pp. 61-78.
[4] D.M. Halliday, J.R. Rosenberg, A.M. Amjad, P. Breeze, B.A. Conway,
and S.F. Farmer, “A Framework for the analysis of mixed time
series/point process data- theory and application to the study of
physiological tremor, single motor unit discharges and
electromyograms,” Prog. Biophys. Molec. Biol.,vol. 64(2-3), 1995,
STFT Coherogram
70 pp.237-278
[5] M.R. Baker, and S.N. Baker, “The effect of diazepam on motor cortical
60 oscillations and corticomuscular coherence studied in man,” J.Physiol,
546(3), 2003, pp.931-942.
50
[6] P. Grosse, M.J. Cassidy, P. Brown, “EEG-EMG, MEG-EMG and EMG-
EMG frequency analysis: physiological principles and clinical
applications,” J.Clinical Neurophysiology, vol. 113, 2002, pp. 1523-
40
Frequency
1531
[7] J.R. Rosenberg, A.M. Amjad, P. Breeze, D.R. Brillinger, and D.M.
30
Halliday, “The Fourier approach to the identification of functional
coupling between neuronal spike trains,” Prog. Biophys. Mol. Biol., vol
20 53, 1989, pp 1-31
[8] S. Mallat, A Wavelets Tour of Signal Processing. San Diego, CA:
10 Academic Press, 1998, ch. 4.
[9] Inc The Math Works, MATLAB wavelets toolbox 2.0 reference
0 guide.
0 5 10 15 20 25 30 35 40 45
Time [10] D.J. Macfarland, L.M. McCane, S.V. David, J.R. Wolpaw, “Spatial filter
selection for EEG-based communication,” Electroencephalogr.Clin.
Fig.8. STFT coherence of the second primary motor cortex channel, Neurophysiol, vol. 103(3), sep 1997, pp.386-394.
with the supplementary motor cortex channel, plotted as an image
in grayscale versus time and frequency. The darker regions indicate
higher levels of coherence.
ix