Escolar Documentos
Profissional Documentos
Cultura Documentos
The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
Author's personal copy
c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861
a r t i c l e i n f o a b s t r a c t
Article history: A recent paper [1] proposed a new technique, termed the channel reactivity-based method
Received 17 August 2012 (CRB), for characterizing EEG alpha rhythms using individual (IAFs) and channel (CAFs) alpha
Received in revised form frequencies. These frequencies were obtained by identifying the frequencies at which the
16 December 2013 power of the alpha rhythms decreases. In the present study, we present a graphical interac-
Accepted 18 December 2013 tive toolbox that can be plugged into the popular open source environment EEGLAB, making
it easy to use CRB. In particular, we illustrate the major functionalities of the software and
Keywords: discuss the advantages of this toolbox for common EEG investigations. The CRB analysis
CRB analysis plugin plugin, along with extended documentation and the sample dataset utilized in this study,
Individual alpha frequency (IAF) is freely available on the web at http://bio.dei.unipd.it/crb/.
Channel alpha frequency (CAF) © 2013 Elsevier Ireland Ltd. All rights reserved.
Reactivity index
Alpha de-synchronization
Event-related de-synchronization
∗
Corresponding author. Tel.: +39 0498277741; fax: +39 0498277699.
E-mail addresses: anahita.goljahani@dei.unipd.it (A. Goljahani), patrizia.bisiacchi@unipd.it (P. Bisiacchi),
gianni@dei.unipd.it (G. Sparacino).
URL: http://www.dei.unipd.it/∼gianni (G. Sparacino).
0169-2607/$ – see front matter © 2013 Elsevier Ireland Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cmpb.2013.12.010
Author's personal copy
854 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861
identifying the alpha frequency intervals instead of using the activity (e.g., the interval of (7.5, 13) Hz in Fig. 1(b)). As illus-
conventional (8, 13) Hz band. Therefore, in this manuscript trated in Fig. 1(c), (fa ,fb ) is expanded leftward until R(f) and
we present an interactive graphical software toolbox, the CRB T(f) intersect and rightward until R(f) and T(f) are close enough
analysis plugin; this set of MATLAB® (The Mathworks Inc.) to consider the alpha modulation activity to be finished. The
routines was designed to be plugged into the popular open required closeness is regulated by a parameter ε. As ε becomes
source platform EEGLAB [12]. increasingly less than 1, R(f) and T(f) must become closer for
The manuscript is organized as follows. First, in Section 2, the alpha interval to be considered finished. In the example
a brief review of CRB is given. Next, in Section 3, the func- of Fig. 1(c), ε = 0.5, and the initial interval (8, 11) Hz is correctly
tionalities of the CRB analysis plugin (e.g., selecting data for expanded to the interval of (7.5, 13) Hz
analysis, applying the method, and inspecting the results) are
described and then illustrated using a sample EEG dataset.
Step 2. Step 2 selects the channels with the strongest alpha
Lastly, in Section 4, possible applications of the software in
de-synchronizations relative to the subject’s alpha activity.
common neuroscientific investigations are discussed.
A reference maximum de-synchronization intensity max is
obtained by computing the pth percentile of the set of val-
2. Review of the CRB method ues computed in Step 1. The CRB parameter p (e.g., 80% or
90%) regulates the sensitivity of the subject’s max to isolated
CRB compares the EEG data of a test state to that of a reference high values. Channels are then selected if their is suffi-
state by means of a power spectral density (PSD) approach ciently close to max . Specifically, the minimum required for
[13,14]; any alpha de-synchronization that may occur when a channel to be selected is given by sub = r·max , where r is the
the subject changes from one state to another is identified. CRB parameter that determines the closeness of the threshold
Fig. 1(a) illustrates the CRB for a sample EEG recorded from sub to max . Before computing max , channels can be discarded
N scalp locations. First, for each channel, the CRB computes independent of the alpha activity of the subject by setting the
the power spectral densities relative to the reference and CRB parameter min to a value greater than 0. In this case, the
test states (denoted as R(f) and T(f), respectively). Then, CRB channels with values smaller than min do not affect the
performs the three steps that are summarized below and determination of max through the percentile operator.
described in detail by Goljahani et al. [1].
Step 3. For all of the selected channels, CAFs are computed
Step 1. For each channel, the alpha de-synchronization as
interval f = (f1 ,f2 ) is identified by applying a “scanning and
expansion” procedure to the corresponding R(f) and T(f). Then, f2
f1
R(f )fdf
the intensity of the alpha de-synchronization in (f1 ,f2 ) is quan- f˛ = f2 , (2)
tified using the reactivity index (V2 /Hz) f1
R(f )df
f2
f1
(R(f ) − T(f ))df where (f1 ,f2 ) are the channel de-synchronization intervals
= (1)
(f2 − f1 ) identified in Step 1. Eq. (2) corresponds to the weighted aver-
age of the frequencies in (f1 ,f2 ), with weights proportional to
which measures the average decrease in power from the ref- the power of those frequencies in the reference state. Lastly,
erence state to the test state for each frequency unit in (f1 ,f2 ). the IAF is computed as the median of the CAFs.
c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861 855
Fig. 1 – Panel (a): block diagram illustrating the three steps of the CRB method. For each channel i, R(f) and T(f) are the resting
and test spectra, respectively. f = (f1 ,f2 ), where f1 and f2 are the boundaries of the alpha modulation interval. is the
reactivity index relative to f. L is the set of channels selected by CRB, and f̂˛ is the estimated IAF. Panels (b) and (c): “scanning
and expansion” procedure illustrated utilizing representative reference (red) and test (blue) EEG spectra. In panel (b), the
sliding windows are represented by rectangles superimposed to the spectra, and the frequency interval selected by the
scanning procedure is highlighted by the thick rectangle. The thin red rectangle is drawn around the frequency interval that
would have been selected in the case of no regularization mechanism. In panel (c), the initial and expanded intervals are
highlighted in dark and light gray, respectively. (For interpretation of the references to color in this figure legend, the reader
is referred to the web version of this article.)
Adapted from Figs. 2, B.1, and B.3 of [1].
Author's personal copy
856 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861
c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861 857
Fig. 2 – Screenshot of the GUI for the initial settings of the CRB analysis toolbox. The GUI contains the initial settings needed
to run the CRB analysis on the sample dataset. Details about the components of the GUI can be found in Section 3.2.2. For
ease of reading, labels A–E were added to the screenshot.
epochs significantly reduces their high single-trial variance. Once all initial settings have been defined, pressing the run
When fewer epochs are available, it may be worthwhile to CRB analysis and show results button computes the results. The
reduce the windows length and increase the overlap values to results are then displayed, as illustrated in the next section.
reduce the variance of the single-trial PSDs at the expense of
a loss in resolution. Accordingly with Welch’s method, data in 3.2.3. Inspection of the results
each window are multiplied by a tapering function. The role The results of the CRB analysis are displayed both numerically,
of the function is to counteract the masking of the lower level as in Fig. 3, and graphically, as in Fig. 4.
components by the sidelobes of the close higher level compo- In addition to the CAFs, s, and boundaries of the de-
nents due to leakage. The default setting of the toolbox is the synchronization intervals (f1 ,f2 ), the numerical results in Fig. 3
Hanning window, but options for the Hamming window or no include a selection flag that is set to 1 for the channels
tapering [21] are also available. Panel D of Fig. 2 defines the that were selected during Step 2 of the algorithm. The CAFs
seven parameters of the CRB algorithm. Although default val- of the selected channels are those that generate the IAF
ues are assigned to the parameters, the user can easily modify reported at the bottom of the window. Analogously, the aver-
these values (several examples that illustrate their role can age modulation interval reported under the IAF is obtained by
be found in Section 5 of the online technical documentation). averaging the boundaries of the de-synchronization intervals
Note that panel D provides two additional options for comput- of the selected channels. The values of the reference, PR , and
ing CAFs in addition to the one defined by Eq. (2). Specifically, test, PT , alpha powers (V2 ) in (f1 ,f2 ) are also included in the
CAFs can be computed by utilizing T(f) or |R(f) − T(f)| instead of table of Fig. 3, along with their relative percentage variation
the R(f) from Eq. (2). In the first case, each CAF reflects the fre- (PT − PR )/PR × 100 [3]. NaN values are returned for the chan-
quencies of the de-synchronized alpha-rhythms with highest nels for which the algorithm was not able to detect any alpha
powers during the test condition. In contrast, in the second de-synchronization. After they are inspected, the results can
case, each CAF reflects the frequencies of the alpha-rhythms either be saved by pressing save results or recomputed after
with the largest decrease in power between the reference and changing the values of the CRB parameters in the panel on
state intervals. the right side of the window.
Author's personal copy
858 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861
Fig. 3 – Screenshot of the GUI for numerical values inspection that results from the CRB analysis of the sample dataset. The
GUI contains a table that reports the following information for each channel: from the left, the selection status, the values of
CAF and , the boundaries (f1 and f2 ) of the modulation interval, reference and test powers in the modulation interval, and
their relative power variation percentage. The value of the IAF is reported in text at the bottom of the GUI. Interactive tools
for refining the initial settings, plotting the spectra of the channels and saving the results are available on the right side of
the GUI.
The de-synchronization intervals, CAFs, selection flags, local average modulation interval are computed using the CAFs
and IAF are also graphically illustrated in the window shown in and the de-synchronization intervals of the selected channels
Fig. 4. For each channel, the reference (red line) and test (blue among those plotted.
line) PSDs are plotted together, and a gray rectangle is drawn
around the de-synchronization interval that was identified by 3.3. Additional interactive tools
the CRB. Moreover, a thin vertical line is drawn along the CAF,
and a check mark in the right upper corner of a plot represents During the inspection of results, the user can easily change
the value of the selection flag for the channel. Lastly, for all the boundaries of (f1 ,f2 ) utilized to compute CAFs, reactiv-
channels, a thick vertical line is drawn at the IAF. The graphical ity indexes and powers. For instance, the user may need
representation of the results allows immediate identification to slightly refine the de-synchronization intervals that were
of the most active scalp regions and frequency intervals of identified by the CRB or may be interested in comparing CRB
the alpha activity. The graphical representation of the results results with those obtained by computing CAFs relative to the
may also reveal the need to refine some of the parameters (e.g., conventional (8, 13) Hz band. New boundaries can be defined
decrease due to the presence of a strong de-synchronization by typing their values in the table of numerical results or by
in the beta range that the algorithm mistakes for an alpha highlighting the desired intervals with the mouse directly on
de-synchronization). If this is the case, the value of can be the plots of Fig. 4. Once the new intervals have been defined,
changed in the window that presents the numerical results, results in the table of Fig. 3 can be updated by pressing the
and the analysis may be re-launched with the new parame- recompute with new table values button in the first case and
ter value. The values of local IAF and local average modulation the synchronize with table data button in the second case. Note
interval at the bottom of the window in Fig. 4 are only different that the power modulations in any other band (e.g., the con-
from those reported at the bottom of the window in Fig. 3 if a ventional (4, 7) Hz theta band) can also be assessed from the
subset of all channels is plotted. In this case, the local IAF and windows in Figs. 3 and 4; this use goes beyond the aim of
Author's personal copy
c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861 859
Fig. 4 – Screenshot of the GUI for spectra plots inspection that results from the CRB analysis of the sample dataset. Each plot in
the GUI depicts the reference PSD (red line) superimposed to the test PSD (blue line) for the channel in the plot title. In all
plots, a thick vertical black line is drawn at the subject’s IAF, which is reported in text at the bottom of the GUI. For all
streams with a positive value, the responsiveness regions are shaded in gray, and a thin vertical black line is drawn along
the relative CAF. The red titles correspond to selected channels. The values of local IAF and the average modulation interval
are displayed at the bottom of the GUI. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)
860 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861
under investigation. Choosing the same setting values allows [6] C.M. Smit, M.J. Wright, N.K. Hansell, G.M. Geffen, N.G.
the results to be reproduced in various laboratories. Moreover, Martin, Genetic variation of individual alpha frequency (IAF)
additional interactive tools allow the user to easily investigate and alpha power in a large adolescent twin sample,
International Journal of Psychophysiology 61 (2006) 235–243.
the frequencies and channel subsets of his/her interest.
[7] C. Babiloni, P. Capotosto, C. Del Percio, F. Babiloni, L. Petrini,
Depending on the initial settings and the outcomes on M. Buttiglione, G. Cibelli, J. Marusiak, G.L. Romani, L.
which the user chooses to focus on, the toolbox can support Arendt-Nielsen, P.M. Rossini, Sensorimotor interaction
diverse investigations. For instance, in addition to studying between somatosensory painful stimuli and motor
IAFs relative to the entire scalp or a scalp sub-region (e.g., [22]), sequences affects both anticipatory alpha rhythms and
the user may focus on CAFs and investigate how their vari- behavior as a function of the event side, Brain Research
ability correlates with cognitive performance or pathological Bulletin 81 (2010)
398–405.
conditions. In addition, users can focus on selection flags and
[8] B.S. Chang, D.L. Schomer, E. Niedermeyer, Normal EEG and
reactivity indexes to determine, e.g., the scalp region at which sleep: adults and elderly, in: D.L. Schomer, F.H. Lopes da
the power is mostly modulated. Lastly, another possible prac- Silva (Eds.), Niedermeyer’s Electroencephalography: Basic
tical application may involve the determination of best scalp Principles, Clinical Applications, and Related Fields, 6th ed.,
locations or frequency intervals for driving brain–computer Lippincott Williams & Wilkins, London, 2011, pp. 183–214.
interfaces (BCIs) based on EEG power modulations (e.g., [9] W. Klimesch, M. Doppelmayr, H. Schimke, B. Ripper, Theta
synchronization and alpha desynchronization in a memory
[23,24]). In conclusion, the CRB analysis plugin is aimed at sup-
task, Psychophysiology 34 (1997) 169–176.
porting investigations into the power-based modulation of EEG
[10] R.H. Grabner, C. Brunner, R. Leeb, C. Neuper, G. Pfurtscheller,
alpha rhythms. To do so, the toolbox combines specific analy- Event-related EEG theta and alpha band oscillatory
sis procedures with visualization tools and thus facilitates the responses during language translation, Brain Research
generation of insights about the investigated phenomenon. Bulletin 72 (2007) 57–65.
The software is open source; thus, further developments can [11] P. Capotosto, M.G. Perrucci, M. Brunetti, C. Del Gratta, M.
easily be included in the toolbox and eventually shared, e.g., to Doppelmayr, R.H. Grabner, W. Klimesch, A. Neubauer, C.
Neuper, G. Pfurtscheller, G.L. Romani, C. Babiloni, Is there
improve its functionalities or to complement frequency analy-
“neural efficiency” during the processing of visuo-spatial
sis with time-domain approaches such as single-trial analysis information in male humans? An EEG study, Behavioural
[25]. Brain Research 205 (2009) 468–474.
[12] A. Delorme, S. Makeig, EEGLAB: an open source toolbox for
analysis of single-trial EEG dynamics including independent
Acknowledgments component analysis, Journal of Neuroscience Methods 134
(2004) 9–21.
The authors thank the anonymous reviewers for the com- [13] A. Brignol, T. Al-ani, X. Drouot, Phase space and power
spectral approaches for EEG-based automatic sleep–wake
ments and criticisms that helped to greatly improve the
classification in humans: a comparative study using short
quality of the final manuscript. The study was partially
and standard epoch lengths, Computer Methods and
supported by the University of Padova “Quantitative under- Programs in Biomedicine 109 (2013)
standing of the human brain functioning through advanced 227–238.
EEG and NIRS signal processing” grant. The authors are grate- [14] B. Yilmaz, S. Korkmaz, D. Betül Arslan, E. Güngör, M.H.
ful to Prof. P. Amodio and Dr. S. Schiff (Department of Clinical Asyal, Like/dislike analysis using EEG: determination of
and Experimental Medicine, University of Padova, Italy), who most discriminative channels and frequencies, Computer
Methods and Programs in Biomedicine (2013),
also collected the data previously published in [1], and to Dr.
http://dx.doi.org/10.1016/j.cmpb.2013.11.010 (in press).
C. D’Avanzo for their earlier contributions to the research that [15] A. Mognon, J. Jovicich, L. Bruzzone, M. Buiatti, ADJUST: an
ultimately led to the development of the software presented automatic EEG artifact detector based on the joint use of
in this study. Part of this work was performed under the aegis spatial and temporal features, Psychophysiology 48 (2011)
of CIRMANMEC, University of Padova. 229–240.
[16] F.C. Viola, J. Thorne, B. Edmonds, T. Schneider, T. Eichele, S.
Debener, Semi-automatic identification of independent
references
components representing EEG artifact, Clinical
Neurophysiology 120 (2009) 868–877.
[17] C.R. Pernet, N. Chauveau, C. Gaspar, G.A. Rousselet, LIMO
[1] A. Goljahani, C. D’Avanzo, S. Schiff, P. Amodio, P. Bisiacchi, G. EEG: a toolbox for hierarchical LInear MOdeling of
Sparacino, A novel method for the determination of the EEG ElectroEncephaloGraphic data, Computational Intelligence
individual alpha frequency, NeuroImage 60 (2012) 774–786. and Neuroscience 2011 (834109) (2011) 1–11.
[2] W.W. Klimesch, EEG alpha and theta oscillations reflect [18] A. Delorme, T. Mullen, C. Kothe, Z. Akalin Acar, N.
cognitive and memory performance : a review and analysis, Bigdely-Shamlo, A. Vankov, S. Makeig, EEGLAB, SIFT, NFT,
Brain Research Reviews 29 (1999) 169–195. BCILAB, and ERICA: new tools for advanced EEG processing,
[3] G. Pfurtscheller, F.H. Lopes da Silva, Event-related EEG/MEG Computational Intelligence and Neuroscience 2011 (130714)
synchronization and desynchronization: basic principles, (2011) 1–12.
Clinical Neurophysiology 110 (1999) 1842–1857. [19] S. Boudet, L. Peyrodie, G. Forzy, A. Pint, H. Toumi, P. Gallois,
[4] C.R. Clark, M.D. Veltmeyer, R.J. Hamilton, E. Simms, R. Paul, Improvements of adaptive filtering by optimal projection to
D. Hermens, E. Gordon, Spontaneous alpha peak frequency filter different artifact types on long duration EEG
predicts working memory performance across the age span, recordings, Computer Methods and Programs in
International Journal of Psychophysiology 53 (2004) 1–9. Biomedicine 108 (2012) 234–249.
[5] P.L. Nunez, R. Srinivasan, Electric Fields of the Brain, 2nd ed., [20] A. Delorme, T. Sejnowski, S. Makeig, Enhanced detection of
Oxford University Press, New York, NY, 2006. artifacts in EEG data using higher-order statistics and
Author's personal copy
c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 853–861 861
independent component analysis, NeuroImage 34 (2007) movements: the simulator, Computer Methods and
1443–1450. Programs in Biomedicine 111 (2013) 300–307.
[21] S.M. Kay, Modern Spectral Estimation: Theory and [24] L. Tonin, R. Leeb, A. Sobolewski, J.R. Millán, An online EEG
Application., Prentice-Hall, Upper Saddle River, NJ, BCI based on covert visuospatial attention in absence of
1988. exogenous stimulation, Journal of Neural Engineering 10
[22] W. Klimesch, H. Schimke, G. Pfurtscheller, Alpha frequency, (2013) 056007.
cognitive load and memory performance, Brain Topography [25] C. D’Avanzo, A. Goljahani, G.L. Pillonetto, G. De Nicolao, G.
5 (1993) 241–251. Sparacino, A multi-task learning approach for the extraction
[23] D.A. Rohani, W.S. Henning, C.E. Thomsen, T.W. Kjaer, S. of single-trial evoked potentials, Computer Methods and
Puthusserypady, H.B.D. Sorensen, BCI using imaginary Programs in Biomedicine 110 (2013) 125–136.