- Email: [email protected]
A new method for quantifying EEG event-related desynchronization: amplitude evvelope analysis
ELSEVIER
Electroencephalography and clinical Neurophysiology 98 ( 1996) 126- 129
A new method for quantifying EEG event-related desynchronization: amplitude envelope analysis Patrice Clochon *, Jean-Marc Fontbonne, Nathalie Lebrun, Pierre Edvenon Laboratoire de Cartographic EEG, U-320 INSERM, Centre Esquirol, CHU Crite de Nacre, 14033 Caen Cedex, France
Accepted for publication: 10 August 1995
Abstract Amplitude modulation (AM) analysis defines precisely the EEG signal envelope changes at sampling frequency. Here we demonstrate mathematically that event-related desynchronization (ERD) corresponds to the integration of AM-EEG. We applied this new approach to a group of 12 healthy human volunteers hearing repeated auditory stimuli and statistically compared the results from ERD to those from AM-EEG. The results indicate that AM-EEG characterized more precisely a specific evoked EEG cortical activation event and may be a powerful method for studying the time-course of functional dynamic brain EEG mapping with improved time resolution. Keywords: Amplitude modulation; EEG; Event-related desynchronization (ERD); Fisher test; Evoked events; Functional brain imaging
1. Introduction The functional brain dynamic study of cognitive or perceptive processes over short time epochs is an important goal presently achieved only by EEG or MEG. With respect to EEG, these events (so-called activations) have been characterized by a reduction of rhythmic components in the alpha or beta frequency bands (Pfurtscheller et al., 1988). Several methods have been developed to analyze and detect such changes. First, quantitative EEG methods have processed spectral changes over long epochs of several seconds or longer. For instance, a decrease in spectral amplitude root mean square (RMS) was considered by Et&enon et al. (1988) as a sign of cortical activation. In order to improve the time resolution, Pfurtscheller and Aranibar (1977) have proposed a method they named “event-related desynchronization” (ERD), which was based on successive 125 msec epochs. An increase in ERD magnitude corresponds to more pronounced desynchronization and can be considered an electrophysiological correlate of cortical activation (Pfurtscheller et al., 1988). With the aim of obtaining a finer time resolution (only restricted by the sampling rate) for the study of cognitive or perceptive processes, we now propose a new method
based on measuring amplitude modulation (AM). To do this, we have applied to the ERD domain the initial theoretical model of EEG complex demodulation (Walter, 19681, extended by Ete’venon et al. (1980), Etbvenon and Giannella (1980) and more recently by Barlow (1993). In order to document the potential of our new method (referred to as AM-EEG), we will first demonstrate that ERD is effectively a time integration of amplitude modulation and then show that AM-EEG produces statistically significant results with shorter time resolution when compared to ERD.
2. Methods 2.1. ERD method ERD computation was performed by filtering the signal in the alpha and beta bands and measuring the RMS amplitude on band power (BP) of 125 msec epochs synchronized by repetitive stimulation. The BP in Eq. 1 is defined by:
BP(n)=/F
(1)
* Corresponding author. Tel.: + 33 31 064439; Fax: +33 31 957272. 0013~4694/96/$15.00 0 I996 Elsevier Science Ireland Ltd. All rights reserved SSDlOOl3-4694(95)00192-l
EEG 947 16
P. Ciochon et al./Electroencephulography
In Eq. 1, n is BP epoch number, p the number of samples in each 125 msec epoch, and k an index of the sum. Mean BP is obtained by averaging across all stimulations. Pfurtscheller et al. (1488) have proposed to standardize the mean BP with respect to a previous base line (BP,,) to obtained ERD: ERD(n) =
BP,,- BP(n)
x lOO[%]
BP
(2)
and clinical Neurophysiology 98 (1996) 126-129
over the 125 msec epochs. Any filtered signal, like EEG, can in fact be written as: EEG(k) = m(k)cos(+(k))
AM-EEG( t) =
m(t) m ref
m,,,
- x lOO[%]
2.3. Analysis of the relationship between ERD and AMEEG
We will demonstrate mat ERD equates with adequate approximation the integration of amplitude modulation
(n+ I)P
d 1
BP(n) =
Our method, based on amplitude modulation, provides a signal envelope. This is only possible if there is no signal outside the frequency band studied. Thus, it is first necessary to filter the signal before computation. Amplitude demodulation is calculated according to the Hilbert Transform (Ete’venon and GiaMella, 1980; Ete’venon et al., 1980; Bhansali and Potter, 1986; Witte et al., 1990; Medl et al., 1992; Barlow, 199311to extract envelope (m(t)> (Fig. 1). AM-EEG is obtained by EEG envelope averaging synchronized by the repetitive stimulations and normalized with a base line cm,,). This normalization, in Eq. 3, respects the variation sense of EEG envelope and is opposite to Pfurtscheller’s ERD definition (Eq. 2).
(4)
In this equation, m(k) is the envelope or amplitude modulation and 4(k) the phase modulation (including carrier frequency), of the signal. Thus, according to the previous Eqs. 1 and 4, BP can be written as:
ref
2.2. Amplitude modulation method
127
p ,z,
m2(k)cos2(+(k))
The analysis of a bandwidth (Bw) filtered signal shows that the amplitude modulation signal cannot have spectral components above the Bw frequency. Consequently, in the alpha band (where Bw = 4 Hz), the signal m(t) may present very slow changes occurring during the 125 msec epochs and can be considered more or less constant (hence, m(k) = M, M being a constant value). In this case, Eq. 5 can be simplified as follows: BP(n) = M
% ;witha=(ns)Pcos2(+(k)) $
(6)
k=np
If the 125 msec integration epoch is greater than the cosine period of Eq. 6, then the value of a becomes approximately equal to p/2, p being the number of samples in each 125 msec epoch. This simplification implies the following straightforward formula: BP(n) = g
(7)
Thus, the ERD is approximately equal to the temporal integration of signal amplitude modulation. There are, however, 2 types of error which are inherent to the ERD
s (0 -
m (0
Fig. 1. Computing of amplitude modulation using Hilbert’s Transform. This consists first (with s(t) signal of 2.56 set sampled at 200 Hz) in computing the Fast Fourier Transform (FFT’) of the EEG filtered signal (s,(t) filtered by Fourier Transform), then in inverting the real and imaginary parts of FFT and finally in obtaining the analytic signal h,(t) through inverse FFT. The envelope m(t) is obtained by computing the modulus of the complex signal (s,(t), h,(t)).
128
P. Clochon et ul./ Electroencephalography and clinical Neurophysiology 98 (1996) 126-129
method with respect to the evaluation of the amplitude mean values. Firstly, there exists an intrinsic error because the amplitude may vary even during the 125 msec epoch. For instance, in the case of fast beta activity, the Bw is greater than 4 Hz and variations in signal amplitude are blurred in 125 msec epochs. Secondly, the model is inadequate because 125 msec epochs, though well adapted for the 8 Hz signal, lead to non-determination for neighboring frequencies such as 7 or 9 Hz (the mean value of the cosine term derives away from l/2). Thus, AM-EEG should be a powerful method to characterize precisely the EEG envelope when compared to ERD which only approximates it.
ERD in Alpha Band 15%
A
-15% I I I I
2.5. Data handling and statistical comparisons A comparison between desynchronization levels (ERD) and envelope decreases (AM-EEG) with respect to the baseline was made for the T3 electrode (corresponding to the left primary auditory area), in the alpha (8.2-12.9 Hz) and beta (12.9-21.5 Hz) frequency bands. Statistics were computed with Fisher’s exact probability test (a non-parametric test, which makes no assumption regarding data distribution) according to Kamisky et al. (1994) and ANOVAs with Bonferroni correction. 3. Results We will present alpha and beta band grand averages for T3 computation of ERDs and AM-EEGs. Regarding ERDs, desynchronizations were first identified by exact probabilities 2 0.05. Further statistical analyses were applied on
AM-EEG
in Alpha Bnnd
Probabilities
in Alpha Band
15 90
B
2.4. Protocol and data acquisition In order to demonstrate the usefulness of AM-EEG and to compare the results obtained with this model with those from ERD in a real-life situation, we performed an experiment on 12 healthy volunteers, all female students, aged 18-2 1 years and exhibiting a clear-cut occipital alpha EEG (> 25 PV peak to peak). Multi-channel EEG signals were recorded according to the international IO-20 system with linked ears reference. Two EOG channels and 1 chin EMG channel were also recorded. The EEG signals were filtered by bandpass filters (0.5-40 Hz) and sampled at 200 Hz (HP workstation). Artifact rejection was made by a neural network (Clochon et al., 1992) and later visually validated by one expert. The auditory stimulus was a tone beep (1760 Hz), presented in a time trapezoidal window defined by 20 msec rise, 180 msec plateau and 20 msec decay. The beep was repeated 80 times every 7 sec. Beep intensity through symmetrical loudspeakers was 65 dB. More than 50 artifact-free stimuli were analyzed by both ERD and AM-EEG for each subject. A baseline epoch (250 msec before stimulation) was used for normalizing ERD and AM-EEG according to Eqs. 2 and 3 respectively.
0
0
- 15 % AM:EEG 0.20
C 0.05 0 I I
AM-EEG
Integration
15 %
D
0
- 15 %
I
I
I
-125
0
125
time (ms)
625
Fig. 2. Analyses of ERD and AM-EEG in the alpha band (8.2- 12.9 Hz). The tone beep stimulus is shown as a black rectangle from 0 to 220 msec. A: several consecutive 125 msec epochs of ERD with associated probabilities; B: the corresponding AM-EEG signal; C: the associated probabilities for AM-EEG. Circled letters “s” and “e” correspond, respectively, to the statistical (P I 0.05) start and end of decreases in envelope amplitude. The circled “m” represents the minimum amplitude of this variations. D: AM-EEG integration over 125 msec epochs to obtain ERD.
AM-EEG data in order to better focalize (“zoom-in”) the start and end of every desynchronized T3 EEG event. With respect to the alpha frequency band (Fig. 2), a desychronization appeared in 3 epochs from 125 to 500 msec post stimulation with an extremum from 250 to 375 msec. A similar but better defined envelope decrease could be detected with AM-EEG, statistically starting 185 msec and ending 525 msec after stimulation. The computed minimum was located at 310 msec and presented a decrease of 10.7% relative to the baseline; the duration of this event was 340 msec. With respect to the beta frequency band (Fig. 3), desynchronization similar to that for the alpha band was obtained. In contrast, the AM-EEG
P. Clochon et al,/
Electroencephalography
ERD in Beta Band
0
- 10% AM-EEG
I
in Bets Band
10 90
_.._ __.__. /.._......... _..I ..
.t----_ ........
98 (1996)
126-129
129
ERD and AM-EEG. We have shown that ERD is approximately equivalent to integrated AM-EEG. Desynchronization may be considered as an AM-EEG or EEG signal envelope decrease. AM-EEG extrema, however, are better defined in time than ERD extrema. This leads to a more detailed time-course definition and resolution for AM-EEG envelope traces and an improvement in the interpretation of results showing electrophysiological phenomena (the relative peak at latency 125 msec in the alpha band) that cannot be displayed by the ERD method. In conclusion, our findings suggest that AM-EEG may be an alternative method to the “classical” ERD method to study time-courses of EEG changes after sensory or cognitive stimulation.
10%
I
and clinical Neurophysiology
i‘..............;
0
~~ - 10% AM-EEG
Probabilities
Acknowledgements
in Beta Band
0.20
This methodological work was part of a DRET Ph.D. grant support to P. Clochon. We are grateful to Dr. J.C. Baron for his constructive remarks.
C 0.05
References
0
,
AM-EEG
9
Integration
10%
D
0
- 10%
I
I
-125
0
125
time (ms)
Fig. 3. Analyses of ERD and AM-EEG in the beta band (12.9-21.5 Display and symbols are the same as in Fig. 2.
625 Hz).
analysis evidenced 2 clear-cut decreases: the first occurred 225-260 msec and the second 295-430 msec after stimulation, with minima locahzed at 240 and 315 msec, and signal amplitude decreases of 7.4% and 7.3%, respectively. Thus, the AM-EEG method may focus fast changes in signal envelope overlooked with the ERD method. Furthermore, additional significant, changes in signal envelope were detected at a probability level of P I 0.05. These changes, which were of generally shorter duration, and hence went unnoticed with the ERD, became evident only with the AM-EEG method.
4. Comment We have demonstrated mathematically, experimentally and statistically that a strong relationship exists between
Barlow, J.S. The Electroencephalogram. Its Patterns and Origins. MIT Press, Cambridge, MA, 1993: 355-379. Bhansali, P.V. and Potter, R. Digital demodulation. IEEE Trans. Instrum. Meas., 1986, IM-35: 324-327. Clochon, P., Perchey, CL, Couque, C., Rebeyrolle, H., Bloyet. D. and Etkvenon, P. Automatized classification by neural networks of EEG signals with artifact rejection. In: 14th Annu. Int. Conf. IEEE Eng. Med. Biol. Sot., Lyon, 1992: 5 l-55. EtCvenon, P. and Giannella, F. Wakiig and sleeping states in rat from an EEG data analyis point of view. Waking Sleeping, 1980, 4: 33-45. Etevenon, P., Giannella, F. and Abamou, F. Modkle d’EEG par modulation radidlectrique d’amplitude et de phase. Rev. EEG Neurophysiol., 1980. 10: 69-80. Etevenon, P., P&on-Magnan, P., Guillou, S., Toussaint, M., Guegen, B., Dcniker, P., Loo, H. and Zarifian, E. A pharmacological model of “local activation.” EEG cartography of caffeine effects in normals. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Huber, Toronto, 1988: 17 1- 178. Kamisky, W., Blair, R.C. and Snider, A.D. An exact statistical method for comparing topographic maps, with any number of subjects and electrodes. Brain Topogr., 1994, 6: 203-210. Medl, A., Flozinger, D. and Pfurtscheller, G. Hilbert-transform based predictions of hand movements from EEG measurements. In: Annu. Int. Conf. IEEE Eng. Med. Biol. Sot., Paris, 1992: 2540. Pfurtscheller, G. and Aranibar, A. Event-related desynchronisation detected by power measurements of scalp EEG. Electroenceph. clin. Neurophysiol., 1977, 42: 817-826. Pfurtscheller, G., Steffan, J. and Maresch, H. ERD mapping and functional topography: temporal and spatial aspects. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Huber, Toronto, 1988: 117-129. Walter, D.O. The method of complex demodulation. In: D.O. Walter and M.A.B. Brazier (Eds.), Advances in EEG Analysis. Electrocnceph. clin. Neurophysiol., Suppl. 27. Elsevier, Amsterdam, 1968: 53-57. Witte, H., Stallknecht, K., Ansorg, J., Griessbach, G., Petranek, S. and Rother, M. Using discrete Hilbert transformation to realize a general methodical basis for dynamic EEG mapping. A methodical investigation. Automedica, 1990, 13: 1- 13.
Electroencephalography and clinical Neurophysiology 98 ( 1996) 126- 129
A new method for quantifying EEG event-related desynchronization: amplitude envelope analysis Patrice Clochon *, Jean-Marc Fontbonne, Nathalie Lebrun, Pierre Edvenon Laboratoire de Cartographic EEG, U-320 INSERM, Centre Esquirol, CHU Crite de Nacre, 14033 Caen Cedex, France
Accepted for publication: 10 August 1995
Abstract Amplitude modulation (AM) analysis defines precisely the EEG signal envelope changes at sampling frequency. Here we demonstrate mathematically that event-related desynchronization (ERD) corresponds to the integration of AM-EEG. We applied this new approach to a group of 12 healthy human volunteers hearing repeated auditory stimuli and statistically compared the results from ERD to those from AM-EEG. The results indicate that AM-EEG characterized more precisely a specific evoked EEG cortical activation event and may be a powerful method for studying the time-course of functional dynamic brain EEG mapping with improved time resolution. Keywords: Amplitude modulation; EEG; Event-related desynchronization (ERD); Fisher test; Evoked events; Functional brain imaging
1. Introduction The functional brain dynamic study of cognitive or perceptive processes over short time epochs is an important goal presently achieved only by EEG or MEG. With respect to EEG, these events (so-called activations) have been characterized by a reduction of rhythmic components in the alpha or beta frequency bands (Pfurtscheller et al., 1988). Several methods have been developed to analyze and detect such changes. First, quantitative EEG methods have processed spectral changes over long epochs of several seconds or longer. For instance, a decrease in spectral amplitude root mean square (RMS) was considered by Et&enon et al. (1988) as a sign of cortical activation. In order to improve the time resolution, Pfurtscheller and Aranibar (1977) have proposed a method they named “event-related desynchronization” (ERD), which was based on successive 125 msec epochs. An increase in ERD magnitude corresponds to more pronounced desynchronization and can be considered an electrophysiological correlate of cortical activation (Pfurtscheller et al., 1988). With the aim of obtaining a finer time resolution (only restricted by the sampling rate) for the study of cognitive or perceptive processes, we now propose a new method
based on measuring amplitude modulation (AM). To do this, we have applied to the ERD domain the initial theoretical model of EEG complex demodulation (Walter, 19681, extended by Ete’venon et al. (1980), Etbvenon and Giannella (1980) and more recently by Barlow (1993). In order to document the potential of our new method (referred to as AM-EEG), we will first demonstrate that ERD is effectively a time integration of amplitude modulation and then show that AM-EEG produces statistically significant results with shorter time resolution when compared to ERD.
2. Methods 2.1. ERD method ERD computation was performed by filtering the signal in the alpha and beta bands and measuring the RMS amplitude on band power (BP) of 125 msec epochs synchronized by repetitive stimulation. The BP in Eq. 1 is defined by:
BP(n)=/F
(1)
* Corresponding author. Tel.: + 33 31 064439; Fax: +33 31 957272. 0013~4694/96/$15.00 0 I996 Elsevier Science Ireland Ltd. All rights reserved SSDlOOl3-4694(95)00192-l
EEG 947 16
P. Ciochon et al./Electroencephulography
In Eq. 1, n is BP epoch number, p the number of samples in each 125 msec epoch, and k an index of the sum. Mean BP is obtained by averaging across all stimulations. Pfurtscheller et al. (1488) have proposed to standardize the mean BP with respect to a previous base line (BP,,) to obtained ERD: ERD(n) =
BP,,- BP(n)
x lOO[%]
BP
(2)
and clinical Neurophysiology 98 (1996) 126-129
over the 125 msec epochs. Any filtered signal, like EEG, can in fact be written as: EEG(k) = m(k)cos(+(k))
AM-EEG( t) =
m(t) m ref
m,,,
- x lOO[%]
2.3. Analysis of the relationship between ERD and AMEEG
We will demonstrate mat ERD equates with adequate approximation the integration of amplitude modulation
(n+ I)P
d 1
BP(n) =
Our method, based on amplitude modulation, provides a signal envelope. This is only possible if there is no signal outside the frequency band studied. Thus, it is first necessary to filter the signal before computation. Amplitude demodulation is calculated according to the Hilbert Transform (Ete’venon and GiaMella, 1980; Ete’venon et al., 1980; Bhansali and Potter, 1986; Witte et al., 1990; Medl et al., 1992; Barlow, 199311to extract envelope (m(t)> (Fig. 1). AM-EEG is obtained by EEG envelope averaging synchronized by the repetitive stimulations and normalized with a base line cm,,). This normalization, in Eq. 3, respects the variation sense of EEG envelope and is opposite to Pfurtscheller’s ERD definition (Eq. 2).
(4)
In this equation, m(k) is the envelope or amplitude modulation and 4(k) the phase modulation (including carrier frequency), of the signal. Thus, according to the previous Eqs. 1 and 4, BP can be written as:
ref
2.2. Amplitude modulation method
127
p ,z,
m2(k)cos2(+(k))
The analysis of a bandwidth (Bw) filtered signal shows that the amplitude modulation signal cannot have spectral components above the Bw frequency. Consequently, in the alpha band (where Bw = 4 Hz), the signal m(t) may present very slow changes occurring during the 125 msec epochs and can be considered more or less constant (hence, m(k) = M, M being a constant value). In this case, Eq. 5 can be simplified as follows: BP(n) = M
% ;witha=(ns)Pcos2(+(k)) $
(6)
k=np
If the 125 msec integration epoch is greater than the cosine period of Eq. 6, then the value of a becomes approximately equal to p/2, p being the number of samples in each 125 msec epoch. This simplification implies the following straightforward formula: BP(n) = g
(7)
Thus, the ERD is approximately equal to the temporal integration of signal amplitude modulation. There are, however, 2 types of error which are inherent to the ERD
s (0 -
m (0
Fig. 1. Computing of amplitude modulation using Hilbert’s Transform. This consists first (with s(t) signal of 2.56 set sampled at 200 Hz) in computing the Fast Fourier Transform (FFT’) of the EEG filtered signal (s,(t) filtered by Fourier Transform), then in inverting the real and imaginary parts of FFT and finally in obtaining the analytic signal h,(t) through inverse FFT. The envelope m(t) is obtained by computing the modulus of the complex signal (s,(t), h,(t)).
128
P. Clochon et ul./ Electroencephalography and clinical Neurophysiology 98 (1996) 126-129
method with respect to the evaluation of the amplitude mean values. Firstly, there exists an intrinsic error because the amplitude may vary even during the 125 msec epoch. For instance, in the case of fast beta activity, the Bw is greater than 4 Hz and variations in signal amplitude are blurred in 125 msec epochs. Secondly, the model is inadequate because 125 msec epochs, though well adapted for the 8 Hz signal, lead to non-determination for neighboring frequencies such as 7 or 9 Hz (the mean value of the cosine term derives away from l/2). Thus, AM-EEG should be a powerful method to characterize precisely the EEG envelope when compared to ERD which only approximates it.
ERD in Alpha Band 15%
A
-15% I I I I
2.5. Data handling and statistical comparisons A comparison between desynchronization levels (ERD) and envelope decreases (AM-EEG) with respect to the baseline was made for the T3 electrode (corresponding to the left primary auditory area), in the alpha (8.2-12.9 Hz) and beta (12.9-21.5 Hz) frequency bands. Statistics were computed with Fisher’s exact probability test (a non-parametric test, which makes no assumption regarding data distribution) according to Kamisky et al. (1994) and ANOVAs with Bonferroni correction. 3. Results We will present alpha and beta band grand averages for T3 computation of ERDs and AM-EEGs. Regarding ERDs, desynchronizations were first identified by exact probabilities 2 0.05. Further statistical analyses were applied on
AM-EEG
in Alpha Bnnd
Probabilities
in Alpha Band
15 90
B
2.4. Protocol and data acquisition In order to demonstrate the usefulness of AM-EEG and to compare the results obtained with this model with those from ERD in a real-life situation, we performed an experiment on 12 healthy volunteers, all female students, aged 18-2 1 years and exhibiting a clear-cut occipital alpha EEG (> 25 PV peak to peak). Multi-channel EEG signals were recorded according to the international IO-20 system with linked ears reference. Two EOG channels and 1 chin EMG channel were also recorded. The EEG signals were filtered by bandpass filters (0.5-40 Hz) and sampled at 200 Hz (HP workstation). Artifact rejection was made by a neural network (Clochon et al., 1992) and later visually validated by one expert. The auditory stimulus was a tone beep (1760 Hz), presented in a time trapezoidal window defined by 20 msec rise, 180 msec plateau and 20 msec decay. The beep was repeated 80 times every 7 sec. Beep intensity through symmetrical loudspeakers was 65 dB. More than 50 artifact-free stimuli were analyzed by both ERD and AM-EEG for each subject. A baseline epoch (250 msec before stimulation) was used for normalizing ERD and AM-EEG according to Eqs. 2 and 3 respectively.
0
0
- 15 % AM:EEG 0.20
C 0.05 0 I I
AM-EEG
Integration
15 %
D
0
- 15 %
I
I
I
-125
0
125
time (ms)
625
Fig. 2. Analyses of ERD and AM-EEG in the alpha band (8.2- 12.9 Hz). The tone beep stimulus is shown as a black rectangle from 0 to 220 msec. A: several consecutive 125 msec epochs of ERD with associated probabilities; B: the corresponding AM-EEG signal; C: the associated probabilities for AM-EEG. Circled letters “s” and “e” correspond, respectively, to the statistical (P I 0.05) start and end of decreases in envelope amplitude. The circled “m” represents the minimum amplitude of this variations. D: AM-EEG integration over 125 msec epochs to obtain ERD.
AM-EEG data in order to better focalize (“zoom-in”) the start and end of every desynchronized T3 EEG event. With respect to the alpha frequency band (Fig. 2), a desychronization appeared in 3 epochs from 125 to 500 msec post stimulation with an extremum from 250 to 375 msec. A similar but better defined envelope decrease could be detected with AM-EEG, statistically starting 185 msec and ending 525 msec after stimulation. The computed minimum was located at 310 msec and presented a decrease of 10.7% relative to the baseline; the duration of this event was 340 msec. With respect to the beta frequency band (Fig. 3), desynchronization similar to that for the alpha band was obtained. In contrast, the AM-EEG
P. Clochon et al,/
Electroencephalography
ERD in Beta Band
0
- 10% AM-EEG
I
in Bets Band
10 90
_.._ __.__. /.._......... _..I ..
.t----_ ........
98 (1996)
126-129
129
ERD and AM-EEG. We have shown that ERD is approximately equivalent to integrated AM-EEG. Desynchronization may be considered as an AM-EEG or EEG signal envelope decrease. AM-EEG extrema, however, are better defined in time than ERD extrema. This leads to a more detailed time-course definition and resolution for AM-EEG envelope traces and an improvement in the interpretation of results showing electrophysiological phenomena (the relative peak at latency 125 msec in the alpha band) that cannot be displayed by the ERD method. In conclusion, our findings suggest that AM-EEG may be an alternative method to the “classical” ERD method to study time-courses of EEG changes after sensory or cognitive stimulation.
10%
I
and clinical Neurophysiology
i‘..............;
0
~~ - 10% AM-EEG
Probabilities
Acknowledgements
in Beta Band
0.20
This methodological work was part of a DRET Ph.D. grant support to P. Clochon. We are grateful to Dr. J.C. Baron for his constructive remarks.
C 0.05
References
0
,
AM-EEG
9
Integration
10%
D
0
- 10%
I
I
-125
0
125
time (ms)
Fig. 3. Analyses of ERD and AM-EEG in the beta band (12.9-21.5 Display and symbols are the same as in Fig. 2.
625 Hz).
analysis evidenced 2 clear-cut decreases: the first occurred 225-260 msec and the second 295-430 msec after stimulation, with minima locahzed at 240 and 315 msec, and signal amplitude decreases of 7.4% and 7.3%, respectively. Thus, the AM-EEG method may focus fast changes in signal envelope overlooked with the ERD method. Furthermore, additional significant, changes in signal envelope were detected at a probability level of P I 0.05. These changes, which were of generally shorter duration, and hence went unnoticed with the ERD, became evident only with the AM-EEG method.
4. Comment We have demonstrated mathematically, experimentally and statistically that a strong relationship exists between
Barlow, J.S. The Electroencephalogram. Its Patterns and Origins. MIT Press, Cambridge, MA, 1993: 355-379. Bhansali, P.V. and Potter, R. Digital demodulation. IEEE Trans. Instrum. Meas., 1986, IM-35: 324-327. Clochon, P., Perchey, CL, Couque, C., Rebeyrolle, H., Bloyet. D. and Etkvenon, P. Automatized classification by neural networks of EEG signals with artifact rejection. In: 14th Annu. Int. Conf. IEEE Eng. Med. Biol. Sot., Lyon, 1992: 5 l-55. EtCvenon, P. and Giannella, F. Wakiig and sleeping states in rat from an EEG data analyis point of view. Waking Sleeping, 1980, 4: 33-45. Etevenon, P., Giannella, F. and Abamou, F. Modkle d’EEG par modulation radidlectrique d’amplitude et de phase. Rev. EEG Neurophysiol., 1980. 10: 69-80. Etevenon, P., P&on-Magnan, P., Guillou, S., Toussaint, M., Guegen, B., Dcniker, P., Loo, H. and Zarifian, E. A pharmacological model of “local activation.” EEG cartography of caffeine effects in normals. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Huber, Toronto, 1988: 17 1- 178. Kamisky, W., Blair, R.C. and Snider, A.D. An exact statistical method for comparing topographic maps, with any number of subjects and electrodes. Brain Topogr., 1994, 6: 203-210. Medl, A., Flozinger, D. and Pfurtscheller, G. Hilbert-transform based predictions of hand movements from EEG measurements. In: Annu. Int. Conf. IEEE Eng. Med. Biol. Sot., Paris, 1992: 2540. Pfurtscheller, G. and Aranibar, A. Event-related desynchronisation detected by power measurements of scalp EEG. Electroenceph. clin. Neurophysiol., 1977, 42: 817-826. Pfurtscheller, G., Steffan, J. and Maresch, H. ERD mapping and functional topography: temporal and spatial aspects. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Huber, Toronto, 1988: 117-129. Walter, D.O. The method of complex demodulation. In: D.O. Walter and M.A.B. Brazier (Eds.), Advances in EEG Analysis. Electrocnceph. clin. Neurophysiol., Suppl. 27. Elsevier, Amsterdam, 1968: 53-57. Witte, H., Stallknecht, K., Ansorg, J., Griessbach, G., Petranek, S. and Rother, M. Using discrete Hilbert transformation to realize a general methodical basis for dynamic EEG mapping. A methodical investigation. Automedica, 1990, 13: 1- 13.