Clinical Neurophysiology xxx (2013) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Clinical Neurophysiology

journal homepage: www.elsevier .com/locate /c l inph

Permutation entropy of scalp EEG: A tool to investigate epilepsiesSuggestions from absence epilepsies

Edoardo Ferlazzo a,b,1, Nadia Mammone c,1, Vittoria Cianci b, Sara Gasparini a,b, Antonio Gambardella a,d,Angelo Labate a, Maria Adele Latella b, Vito Sofia e, Maurizio Elia f, Francesco Carlo Morabito c,Umberto Aguglia a,b,⇑a Magna Græcia University, Catanzaro, Italyb Regional Epilepsy Centre, ‘‘Bianchi-Melacrino-Morelli’’ Hospital, Reggio Calabria, Italyc University Mediterranea of Reggio Calabria, Italyd National Research Council, Piano Lago di Mangone, Cosenza, Italye Institute of Neurology, University of Catania, Italyf Unit of Neurology and Clinical Neurophysiopathology, Oasi Institute for Research on Mental Retardation and Brain Aging (IRCCS), Troina, Enna, Italy

a r t i c l e i n f o

Article history:Accepted 21 June 2013Available online xxxx

Keywords:Permutation entropyAbsencesEEGBrain topography

/$36.00 � 2013 International Federation of Clinical Nhttp://dx.doi.org/10.1016/j.clinph.2013.06.023

⇑ Corresponding author at: Regional Epilepsy Centrli’’ Hospital, Reggio Calabria, Italy. Tel.: +39 09653979

E-mail address: u.aguglia@tin.it (U. Aguglia).1 These authors contributed equally to this work.

Please cite this article in press as: Ferlazzo E etj.clinph.2013.06.023

h i g h l i g h t s

� We evaluated permutation entropy (PE) extracted from different electrodes in patients with typicalabsences (TAs) and healthy subjects.

� In patients with TAs there was a recurrent behavior of PE topography, with electrodes from anteriorregions associated to higher randomness and electrodes from posterior regions associated to lowerrandomness.

� PE seems a useful tool to disclose abnormalities of cerebral electric activity not revealed by conven-tional EEG.

a b s t r a c t

Objective: We used permutation entropy (PE) to disclose abnormalities of cerebral activity in patientswith typical absences (TAs).Methods: We evaluated 24 EEG of TA patients and 40 EEG of healthy subjects. PE was estimated channelby channel, with electrodes being divided into high-PE cluster (high randomness), low-PE cluster (lowrandomness), and neutral cluster. We compared PE between EEG of patients and controls, and betweeninterictal and ictal EEG of patients.Results: Patients showed a recurrent behavior of PE topography, with anterior brain regions constantlyassociated to high PE levels and posterior brain regions constantly associated to low PE levels, duringboth interictal and ictal phases. On the contrary, healthy controls had a random distribution of PE topog-raphy.Conclusions: In patients with TAs, a higher randomness in fronto-temporal areas and a lower randomnessin posterior areas occur during both interictal and ictal phases. Such abnormalities are in keeping withevidences from different morphological and functional studies showing multifocal brain changes in TApatients.Significance: PE seems to be a useful tool to disclose abnormalities of cerebral electric activity notrevealed by conventional EEG recordings, opening interesting prospective for future studies.

� 2013 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland All rightsreserved.

europhysiology. Published by Elsev

e, ‘‘Bianchi-Melacrino-Morel-72; fax: +39 0965397973.

al. Permutation entropy of sca

1. Introduction

Permutation Entropy (PE) represents a recently introduced wayto assess the complexity of non-linear (chaotic) time series. PE hassome peculiar advantages as compared to other commonly usedentropy measures (such as Approximate Entropy), since it is an

ier Ireland All rights reserved.

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

2 E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx

ordinal measure (Bandt and Pompe, 2002). Indeed, PE decomposesthe original EEG series into a series of ordinal patterns describingthe order relations between the present and a fixed number ofequidistant past values at a given time. To compute PE, the EEGchannel time series do not need to be normalized. PE is able tocatch the differences between random and regular behavior ofcerebral electrical activity, to track transient dynamics of EEGrecordings (Bandt and Pompe, 2002) and to distinguish betweeninterictal and ictal phases (Li et al., 2007). PE has also been usedto predict absence seizures in rats, to differentiate sleep and wak-ing state and to quantify the hypnotic brain effects of general anes-thetic drugs (Li et al., 2007; Bruzzo et al., 2008; Olofsen et al.,2008). PE has been shown to yield a relatively easy way to charac-terize the properties of regularity (complexity) of a single channeltime-series. Among the various nonlinear approaches proposed inthe literature (e.g., Lyapunov Exponents, Approximate Entropy),PE has the advantage to be easy to compute also in a short timewindow, since it does not rely on asymptotic properties but juston a symbolic description of the original time signal.

In this study, we looked at the behavior of the PE extracted fromdifferent electrodes over time in patients with typical absences(TAs). TAs are brief (5–30 s) epileptic seizures of sudden onsetand termination clinically characterized by more or less severeimpairment of consciousness at times associated with other clini-cal manifestations. TAs may occur in several idiopathic generalizedepilepsies (IGE) including Childhood Absence Epilepsy (CAE) andJuvenile Absence Epilepsy (JAE). The pathophysiological mecha-nisms of TAs have been studied both in human and animal models.Several studies showed that TAs might not be truly generalizedfrom the beginning since a focal focus may be the true generatorof the discharge quickly generalizing through thalamo-corticalpathways (Meeren et al., 2005). Microdysgenesis and other cere-bral structural or functional focal changes have been reported inpatients with IGE from autopsy (Meencke and Janz, 1984), neuro-physiological (Holmes et al., 2004; Westmijse et al., 2009) and ad-vanced neuroimaging studies (Woermann et al., 1998, 1999; Koeppand Duncan, 2000; Savic et al., 2000; Caplan et al., 2009; Moelleret al., 2010). Interictal EEG in patients TAs may be normal or showdiffuse, short-lasting spike-and-wave discharges, occipital inter-mittent rhythmic delta activity (OIRDA), only rarely focal abnor-malities (Panayiotopoulos et al., 1989; Sadleir et al., 2006). IctalEEG of TAs is easily recognized, showing bilateral, regular and sym-metrical, 3 Hz spike-and-wave complexes and it has been studiedby conventional EEG, including brain topography (Silva et al.,1999; Clemens et al., 2000; Vuilleumier et al., 2000; Kim et al.,2011). Conventional brain topography allows colored maps of dif-ferent EEG activities to be created (Harner, 1986). Herein, we per-formed a novel brain topography technique based on PE extractedfrom different electrodes over time, following the temporal evolu-tion of the EEG within a time interval. Thus, ‘‘time’’ variable is ta-ken into account in PE computation (Bandt and Pompe, 2002) asalready done with different complexity measures (Mammoneet al., 2006, 2010). In particular, PE of interictal EEG of patientswith TAs was compared with PE of EEG from healthy subjects todisclose abnormalities of interictal cerebral activity not revealedby conventional EEG recordings. We also compared PE of ictaland interictal EEGs to detect PE changes between these two differ-ent phases in the patients.

2. Methods

2.1. Subjects and EEG recordings

In this study, we analyzed 24 EEG recordings from 24 patients(11 males, 13 females; average age 7.5 years; range 6–14 years) af-

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

fected by CAE (17 patients) or JAE (7 patients), and 40 EEG record-ings from 40 healthy volunteers (18 males, 22 females; average age10.5; range 6–18 years). Twenty out of 24 patients were on val-proic acid, while 4 were receiving no treatment. No control wason medication. A total of 52 TAs were recorded and analyzed. EEGswere recorded with scalp electrodes placed according to the Inter-national 10–20 system with referential montages using G2 (be-tween electrodes Fz and Cz) as the reference. All patients hadundergone routine waking EEG usually lasting around 20 min,including activating procedures (i.e., hyperventilation and inter-mittent light stimulation). The EEG was high pass filtered at0.5 Hz, low-pass filtered at 70 Hz, plus a 50 Hz notch filter. Thesampling rate was set at 256 Hz. After acquisition, each EEG chan-nel was processed as detailed in the following Section.

2.2. Permutation entropy estimation

The EEG was buffered in 10-s overlapping windows, with 1 sshift forward, namely with 9-s overlap. This was implementedby selecting an ideal 10 s time window over the EEG, by estimat-ing PE for each channel within that window, then making thewindow slide 1 s forward over the EEG until the whole tracingis scanned and processed. Fig. 1 shows a sliding window scan-ning the EEG and computing the corresponding PE samples, win-dow by window. Since we estimated one PE sample per channelfor each window, PE value at time t was calculated from thedata between t and (t � 10) s, taking into account the ‘‘time’’ var-iable (see the two following equations). EEG was then convertedin an output compressed tracing (with the same number ofchannels but a reduced number of samples) representing thePE trend of the different EEG channels. We named the singleEEG channel as ‘‘x’’. To calculate PE, each EEG time series ‘‘x’’was mapped into a m-dimensional space, with m being theembedding dimension and L being the time lag. Vectors ‘‘Xt’’were constructed picking up ‘‘m’’ equally spaced samples fromx, starting from time point ‘‘t’’:

Xt ¼ ½xðtÞ; xðt þ LÞ; . . . ; xðt þ ðm� 1ÞL�T

where T indicated transpose.Every vector Xt was then arranged in an increasing order and re-

named Xrt:

Xrt ¼ ½xðt þ ðt1 � 1ÞLÞ; xðt þ ðt2 � 1ÞLÞ; . . . ; xðt þ ðtm � 1ÞLÞ�T

Therefore, each vector Xt could be considered uniquely mappedonto a symbol vector p = [t1, t2, ..., tm]. The vector p is a sequence oftime points, hence a symbol. The computation procedure implies amapping of the original time series onto a sequence of symbolsthat are independent from EEG signal amplitude. In turn, thisimplies that no preprocessing like normalization is required.Pre-normalization is commonly used in amplitude-dependentmethodologies leading to unavoidable lost of information. There-fore, skipping pre-normalization would save part of the informa-tion embedded in the EEG. The frequency of occurrence of eachpossible pi is indicated as p(pi), that represents the number oftimes the specific vector pi is present in the considered time win-dow, normalized by N � (m � 1)L (therefore p (pi) = p (pi)/(N � (m � 1)L)) where N is the number of samples of the timeseries x.

PE is finally defined as:

HðmÞ ¼ �Xm!

i¼1

pðpiÞ ln pðpiÞ

where ln is the natural logarithm and m! is the number of the pos-sible permutations.

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

Fig. 1. PE estimation from EEG. The sliding window procedure. Following acquisition, a virtual 10 s time window is made running over the EEG, estimating PE for eachchannel within the window. The window slides 1 s forward over the EEG until the whole tracing is scanned and processed. PE is estimated within each window, channel bychannel, thus generating a new multichannel dataset showing the PE trend.

Fig. 2. An explanatory plot of an EEG and the corresponding PE. The EEG (channel Fp1) of a patient with TAs is depicted in the top subplot whereas the PE extracted from thesame channel is depicted in the bottom subplot.

E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx 3

Since H(m) can maximally reach ln(m!), PE is generally normal-ized as:

HnðmÞ ¼ �

Xm!

i¼1

pðpiÞ ln pðpiÞ

lnðm!Þ

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

so that its lower bound is 0 and its upper bound is 1. PE is a measureof regularity of the time series: in case of a completely random timeseries, all possible permutations (symbol vector p) occur with thesame probability, whereas, in the case of a time series showing acertain regularity, some permutation vectors will occur with a high-er probability, then Hn(m) decreases. In the present work, the

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

Fig. 3. PE trend (channel Fp1) in the preictal-ictal phase for each patient in the dataset. Each subplot corresponds to a different patient and shows a time window covering200 s of a peri-ictal phase. The seizure is centered at 130 s so that the same temporal segment for every patient is shown. Vertical dotted lines indicate seizure onset.

4 E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx

embedding dimension is fixed to m = 3, and the time lag can assumetwo different values L = 1 and L = 2 (Olofsen et al., 2008). After PEprofiles were estimated for each channel, they were arranged asrows of a matrix (named matrix X hereinafter) whose column vec-tor X(t) represents the set of the PE values associated with the ‘‘n’’electrodes at a certain time window ‘‘t’’. An explanatory plot of theEEG trend and the extracted PE is depicted in Fig. 2.

2.3. Plotting the trend of PE

PE estimation was applied to the analysis of the 24 patients toplot their PE trends and check the ability of this descriptor to re-flect the increased regularity of the EEG during the ictal state. Tovalidate the choice of PE as proper candidate to disclose ictalevents in the EEG of patients with TAs, we evaluated how PE mod-ified during the whole EEG recordings. We observed that PE clearlystarts dropping a few seconds before seizure onset and appears re-duced during ictal phase in all patients (Fig. 3), thus endorsing thehypothesis that EEG signal becomes more regular during ictalphase. Indeed, PE is known to be sensitive to the frequency ofEEG waves (Olofsen et al., 2008) (see Supplementary Figure S1for an example of comparison between a conventional method incontrast to PE analysis). Since entropy spatio-temporal topographyshowed to be more informative than the mere temporal visualiza-tion of entropy profiles (Mammone et al., 2011), we decided to ap-

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

ply this kind of representation to PE. This allowed us to visualizethe temporal evolution of PE topography through a movie. Fig. 4shows an explanatory sequence of frames (see also SupplementaryMovie S1). To automatically extract the spatio-temporal informa-tion about PE topography, we performed a spatio-temporal cluster-ing of the electrodes according to their PE levels. Again, a slidingwindow procedure was used, scanning the PE tracings instead ofthe EEG recordings. Within each temporal window, the electrodeswere clustered in 3 PE groups (high-PE, low-PE and average-PE).The high-PE cluster of electrodes would reflect underlying rela-tively higher random activity, whereas the low-PE cluster of elec-trodes would reflect relatively lower random activity. Average-PEcluster would account for ‘‘neutral’’ activity and was not consid-ered of particular interest. The width of the sliding window wasset at 1 min, therefore, each window included 60 PE sample vectorsX(t) (each sample vector represents the set of the PE values associ-ated with the ‘‘n’’ electrodes at a certain window ‘‘t’’). To cluster thePE groups of electrodes, a clustering technique was necessary andk-means algorithm was selected being a powerful and simple unsu-pervised learning algorithm that solves the clustering problem(Lloyd, 1982). Each PE window was processed by k-means algo-rithm. Once the electrodes were clustered, the algorithm estimatedhow often (time% with respect to the overall length of the EEGrecording) each electrode belonged to one of the three PE cluster.We then came up with a ‘‘membership rate’’ of each electrode to

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

Fig. 4. An explanatory sequence of frames (30 s countdown to the seizure and 10 s afterwards) extracted from one of the movies. Each map represents a frame of the movie ofPE topography evolution. The 2D map with PE topography at the corresponding time point ‘‘t’’ is shown. The plot exploits a function of the toolbox EEGLAB (Delorme andMakeig, 2004). PE levels are encoded with a coloration going from blue (low-PE) to red (high-PE). The color of the spatial points lying between the electrodes are calculated byinterpolation, thus a smooth gradation of colors is achieved. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of thisarticle.)

E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx 5

the three different clusters, finally evaluating if significant differ-ences between the whole EEG of patients and the EEG of controls,the interictal EEG of patients and the EEG of controls, the interictaland ictal EEG of patients, do exist. Statistical analysis was per-formed by mean of the Mann–Whitney U test. Differences wereconsidered statistically significant at p < 0.05. All calculations weredone using a standard statistical package (SPSS for Windows, ver-sion 9.0.1). More details of the statistical analysis procedure are gi-ven in the Supplementary Appendix.

Fig. 5. Box plots of the membership rates to the High-PE cluster of anteriorelectrodes for the four considered cases: Whole patients’ EEG (WP-EEG), ControlEEG (C-EEG), Interictal-EEG and Ictal-EEG. On each box, the central mark is themedian, the edges of the box are the 25th and 75th percentiles, the distancebetween the 25th and the 75th percentiles represents the interquartile range; thewhiskers extend to the most extreme data points not considered outliers.

3. Results

PE changes during interictal and ictal phases were firstly appre-ciated by reviewing the movies of PE mapping evolution (Supple-mentary Movie S1). We observed that electrodes from anteriorbrain areas (Fp1, Fp2, F3, F4, F7, F8, C3, C4, T3, T4) were associatedto relative high-PE whereas electrodes from posterior brain areas(P3, P4, T5, T6, O1, O2) were associated to relative low-PE. A typicalsequence of frames (30 s countdown to the seizure and 10 s after-wards) extracted from the explanatory movie (Supplementary Mo-vie S1) is depicted in Fig. 4. Of note, PE was distributed in a randomfashion in control subjects.

Such observations were thus quantified according to the proce-dure described in methods. By comparing electrodes membership

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scalp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/j.clinph.2013.06.023

Fig. 6. Box plots of the membership rates to the Low-PE cluster of the posteriorelectrodes for the four considered cases: Whole Patients’ EEG (WP-EEG), ControlEEG (C-EEG), Interictal-EEG and Ictal-EEG. On each box, the central mark is themedian, the edges of the box are the 25th and 75th percentiles, the distancebetween the 25th and the 75th percentiles represents the interquartile range; thewhiskers extend to the most extreme data points not considered outliers.

6 E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx

rate in the whole EEG recordings between patients and controls,we observed that electrodes from anterior brain areas (Fp1, Fp2,F3, F4, F7, F8, C3, C4, T3, T4) had a higher membership rate tothe high-PE cluster in patients (mean time% = 71.23, SD 14.56) than

Fig. 7. Comparison between the average PE of the group of anterior electrodes and the grohold PE values that remain steadily higher as compared to posterior electrodes. In the ictelectrodes.

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

in controls (mean time% = 64.42, SD 16.14; p = 0.0014) (Fig. 5),whereas electrodes from posterior brain areas (P3, P4, T5, T6, O1,O2) had a higher membership rate to low-PE cluster in patients(mean time% = 92.08, SD 5.91) than in controls (meantime% = 68.65, SD 22.28; p = 0.0029) (Fig. 6).

By comparing electrode membership rate between PE of inter-ictal EEG of patients and EEG of controls, it emerged that electrodesfrom anterior brain areas had a higher membership rate to thehigh-PE cluster in patients (mean time% = 71.22; SD 14.44) thanin controls (mean time% = 64.42, SD 16.14; p = 0.0015) (Fig. 5),whereas electrodes from posterior brain areas had a higher mem-bership rate to low-PE cluster in patients (mean time% = 92.07, SD5.9) than in controls (mean time% = 68.65, SD 22.28; p = 0.0028)(Fig. 6).

Although PE dropped in all electrodes during ictal phase(Figs. 2–4 and 8), both anterior and posterior electrodes main-tained the same membership rate to high-PE cluster and to low-PE cluster respectively, during both interictal and ictal phases(Figs. 5 and 6). A comparison between the average PE of the groupof anterior and posterior electrodes in EEG recordings from pa-tients and controls is well depicted in Figs. 7 and 8.

4. Discussion

In this study we showed that patients with TAs had a recur-rent behavior of PE topography. In particular, electrodes fromanterior brain regions appear constantly associated to relativehigher PE levels (higher randomness) while those from posteriorbrain regions appear constantly associated to relative low-PE

up of posterior electrodes in a patient’s EEG recording. Note that anterior electrodesal phase, PE values drops sharply, even though they still maintain higher in anterior

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

Fig. 8. Comparison between the average PE of the group of anterior electrodes and the group of posterior electrodes of a control subject’s EEG. Please note that the average PEof the two groups of electrodes overlaps throughout the whole recording.

E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx 7

(lower randomness). On the contrary, healthy controls show arandom distribution of PE topography. We also demonstrated thatelectrodes belonging to anterior regions and posterior regions re-sult clustered (higher randomness in fronto-temporal areas andlower randomness in posterior areas) during both interictal andictal phases. We may hypothesize that the imbalance of EEG com-plexity between anterior and posterior brain regions in IGE pa-tients with TAs may be due to an abnormal functionalconnectivity among these areas. Our findings are in agreementwith Koutroumanidis et al. (2012), who performed a topographicanalysis of focal and generalized spike and wave discharges in pa-tients with TAs, finally proposing that a system of multifocal(mainly anterior and posterior) cortical areas are reciprocallyinterrelated between them and the thalamic sectors, operatingas a system. Focal, multifocal and diffuse structural brain changeshave also been demonstrated by other methods. A neuropatho-logical study (Meencke and Janz, 1984) showed that seven ofeight patients with IGE had multiple foci of microdysgenesis. Ad-vanced neuroimaging studies in patients with IGE showed signif-icantly larger cortical grey matter volumes with increasedcortical/subcortical volume ratio compared with controls (Woer-mann et al., 1998). These authors postulated that such findingswere related to abnormal neuronal connectivity. The same group(Woermann et al., 1999) also found an increase in cortical greymatter in mesial frontal lobes in many patients with JME, sup-porting the concept of regional structural changes associated withabnormalities of functional connectivity in these patients. PETstudies also revealed localized metabolic abnormalities in the me-sial frontal region in IGE (Koepp and Duncan, 2000). Savic et al.(2000) evaluated JME patients through MR spectroscopy andfound reduced concentration of N-acetyl aspartate in the mesio-frontal cortex suggesting a frontal lobe dysfunction. VolumetricMRI studies demonstrated significantly smaller gray matter vol-umes of the left orbital frontal gyrus and both temporal lobesin children with CAE as compared to children without epilepsy

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

(Caplan et al., 2009). Additional evidence that the cortical mecha-nisms of IGE are complex and non-uniform is given by generalizedreflex seizures (Ferlazzo et al., 2005). Indeed, many patients withphotosensitivity or pattern sensitivity, as well as seizures inducedby verbal and non-verbal cognitive stimuli, have IGE. Functionalimaging and other methods have contributed to support the hypoth-esis that patients with IGE may present complex hyperexcitable net-works extending over multiple cortical areas that, when receivingappropriate afferent volleys, may finally produce an epileptic activ-ity (Ferlazzo et al., 2005). Finally, the presence of ictal and intercitalmultifocal (mainly frontal or occipital) abnormalities in EEG of pa-tients with CAE (Koutroumanidis et al., 2012) further supports thehypothesis that hyperexitability may not be uniformly distributedover both hemispheres in such conditions.

PE has been used to predict absence seizure in rats (Li et al.,2007) and to differentiate sleep and waking state (Bruzzo et al.,2008). Li et al. (2007) evaluated PE in genetic absence epilepsy ratsby recording EEG activity through one intracerebral electrode im-planted in frontal cortex. They showed that PE gradually decreasedbefore seizures, and dropped in ictal phase, suggesting that PEcould predict seizure onset. Of course, by using one electrode theycould not evaluate how PE varies in different brain regions duringinterictal or ictal stages. Our study confirms and further expandstheir findings in human epileptic subjects, showing that PE levelsdrop everywhere in ictal phase, remaining relatively higher inanterior electrodes as compared to posterior ones. Bruzzo et al.(2008) evaluated six seizures in a population different from ours(drug-resistant focal epilepsy), showing that PE may change withmodification of vigilance state. In an EEG-fMRI study on patientswith IGE (Benuzzi et al., 2012), hemodynamic changes precedinggeneralized spike and wave discharges were found in some areasbelonging to the so called ‘‘default mode network’’ (DMN) (Raichleet al., 2001), that is an organized brain function presents at rest andsuspended during specific goal-directed behavior. Authors postu-lated that such changes could be related to physiological alertness

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/

8 E. Ferlazzo et al. / Clinical Neurophysiology xxx (2013) xxx–xxx

oscillations. In our sample, TAs were recorded during wakefulness,often during hyperventilation with breath counting, thus being ex-tremely unlikely that PE changes may be due to low level of vigi-lance. Hence, we can confirm that PE may drop in all electrodesin preictal and ictal stages in waking states. Benuzzi et al. (2012)also suggested that those hemodynamic changes preceding gener-alized spike and wave discharges could be due to a synchronizationof neuronal activity in ‘‘dysfunctional’’ areas belonging to DMN(frontal cortices, precuneus/posterior cingulate, lateral parietal).This theory is in keeping with our results showing that anteriorand posterior brain regions are steadily clustered, and it stronglysupports the hypothesis that an imbalanced functional connectiv-ity between these regions occurs in TAs patients.

In conclusion, PE-mapping seems to be a useful tool for disclos-ing abnormalities of cerebral electric activity not revealed by con-ventional EEG recordings, opening interesting prospective forfuture studies.

Financial interests

None.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.clinph.2013.06.023.

References

Bandt C, Pompe B. Permutation entropy: a natural complexity measure for timeseries. Phys Rev Lett 2002;88:174102.

Benuzzi F, Mirandola L, Pugnaghi M, Farinelli V, Tassinari CA, Capovilla G, et al.Increased cortical BOLD signal anticipates generalized spike and wavedischarges in adolescents and adults with idiopathic generalized epilepsies.Epilepsia 2012;53:622–30.

Bruzzo A, Gesierisch B, Santi M, Tassinari CA, Birbaumer N, Rubboli G. Permutationentropy to detect vigilance changes and preictal states from scalp EEG inepileptic patients. A preliminary study. Neurol Sci 2008;29:3–9.

Caplan R, Levitt J, Siddarth P, Wu KN, Gurbani S, Sankar R, et al. Frontal andtemporal volumes in childhood absence epilepsy. Epilepsia 2009;50:2466–72.

Clemens B, Szigeti G, Barta Z. EEG frequency profiles of idiopathic generalizedepilepsy syndromes. Epilepsy Res 2000;42:105–15.

Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trialEEG dynamics including independent component analysis. J Neurosci Methods2004;134:9–21.

Ferlazzo E, Zifkin BG, Andermann E, Andermann F. Cortical triggers in generalizedreflex seizures and epilepsies. Brain 2005;128:700–10.

Harner RN. Clinical application of computed EEG topography. In: Duffy Fh, editor.Topographic mapping of brain electrical activity. Boston: Butterworths; 1986. p.347–56.

Please cite this article in press as: Ferlazzo E et al. Permutation entropy of scaj.clinph.2013.06.023

Holmes MD, Brown M, Tucker DM. Are ‘‘generalized’’ seizures truly generalized?Evidence of localized mesial frontal and frontopolar discharges in absence.Epilepsia 2004;45:1568–79.

Kim DS, Nordli Jr DR, Zelko F. Spectral power of 1–4 Hz frequency in the ictal phaseof childhood absence epilepsy. J Clin Neurophysiol 2011;28:463–8.

Koepp M, Duncan J. Positron emission tomography in idiopathic generalizedepilepsy: imaging beyond structure. In: Schmidt D, editor. Juvenile myoclonicepilepsy: the Janz syndrome. Petersfield: Rightson Biomedical Publications;2000. p. 91–100.

Koutroumanidis M, Tsiptsios D, Kokkinos V, Kostopoulos GK. Focal and generalizedEEG paroxysms in childhood absence epilepsy: topographic associations anddistinctive behaviors during the first cycle of non-REM sleep. Epilepsia2012;53:840–9.

Li X, Ouyang G, Richards DA. Predictability analysis of absence seizures withpermutation entropy. Epilepsy Res 2007;77:70–4.

Lloyd SP. Least squares quantization in PCM. IEEE Trans Inf Theory 1982;28:129–37.Mammone N, Morabito FC, Principe JC. Visualization of the Short Term Maximum

Lyapunov Exponent Topography in the Epileptic Brain. In: Proceedings of the28th IEEE EMBS annual international conference (EMBC) 2006;1:4257–4260.

Mammone N, Principe JC, Morabito FC, Shiau DS, Sackellares JC. Visualization andmodelling of STLmax topographic brain activity maps. J Neurosci Methods2010;189:281–94.

Mammone N, Inuso G, La Foresta F, Versaci M, Morabito FC. Clustering of entropytopography in epileptic electroencephalography. Neural Comput Appl2011;20:825–33.

Meencke HJ, Janz D. Neuropathological findings in primary generalized epilepsy: astudy of eight cases. Epilepsia 1984;25:8–21.

Meeren H, van Luijtelaar G, Lopes da Silva F, Coenen A. Evolving concepts on thepathophysiology of absence seizures: the cortical focus theory. Arch Neurol2005;62:371–6.

Moeller F, LeVan P, Muhle H, Stephani U, Dubeau F, Siniatchkin M, et al. Absenceseizures: individual patterns revealed by EEG-fMRI. Epilepsia2010;51:2000–10.

Olofsen E, Sleigh JW, Dahan A. Permutation entropy of the electroencephalogram: ameasure of anaesthetic drug effect. Br J Anaesth 2008;101:810–21.

Panayiotopoulos CP, Obeid T, Waheed G. Differentiation of typical absence seizuresin epileptic syndromes. A video EEG study of 224 seizures in 20 patients. Brain1989;112:1039–56.

Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL. Adefault mode of brain function. Proc Natl Acad Sci U S A 2001;98:676–82.

Sadleir LG, Farrel K, Smith S, Connolly MB, Scheffer IE. Electroclinical features ofabsence seizures in childhood absence epilepsy. Neurology 2006;67:413–8.

Savic I, Lekvall A, Greitz D, Helms G. MR spectroscopy shows reduced frontal lobeconcentrations of N-acetyl aspartate in patients with juvenile myoclonicepilepsy. Epilepsia 2000;41:290–6.

Silva C, Pimentel IR, Andrade A, Foreid JP, Ducla-Soares E. Correlation dimensionmaps of EEG from epileptic absences. Brain Topogr 1999;11:201–9.

Vuilleumier P, Assal F, Blanke O, Jallom P. Distinct behavioral and EEG topographiccorrelates of loss of consciousness in absences. Epilepsia 2000;41:687–93.

Westmijse I, Ossenblok P, Gunning B, van Luijtelaar G. Onset and propagation ofspike and slow wave discharges in human absence epilepsy: a MEG study.Epilepsia 2009;50:2538–48.

Woermann FG, Sisodiya SM, Free SL, Duncan JS. Quantitative MRI in patients withidiopathic generalized epilepsy. Evidence of widespread cerebral structuralchanges. Brain 1998;121:1661–7.

Woermann FG, Free SL, Koepp MJ, Sisodiya SM, Duncan JS. Abnormal cerebralstructure in juvenile myoclonic epilepsy demonstrated with voxel-basedanalysis of MRI. Brain 1999;122:2101–7.

lp EEG: A tool to investigate epilepsies. (2013), http://dx.doi.org/10.1016/