The effect of age on human motor electrocorticographic signals and implications for

brainndashcomputer interface applications

This article has been downloaded from IOPscience Please scroll down to see the full text article

2011 J Neural Eng 8 046013

(httpiopscienceioporg1741-255284046013)

Download details

IP Address 128252210217

The article was downloaded on 04022012 at 1916

Please note that terms and conditions apply

View the table of contents for this issue or go to the journal homepage for more

Home Search Collections Journals About Contact us My IOPscience

IOP PUBLISHING JOURNAL OF NEURAL ENGINEERING

J Neural Eng 8 (2011) 046013 (10pp) doi1010881741-256084046013

The effect of age on human motorelectrocorticographic signals andimplications for brainndashcomputerinterface applicationsJarod Roland1 Kai Miller2 Zac Freudenburg3 Mohit Sharma4Matthew Smyth1 Charles Gaona4 Jonathan Breshears1Maurizio Corbetta5 and Eric C Leuthardt146

1 Department of Neurological Surgery Washington University School of Medicine St Louis MO USA2 Department of Physics University of Washington Seattle WA USA3 Department of Computer Science Washington University St Louis MO USA4 Department of Biomedical Engineering Washington University in St Louis St Louis MO USA5 Department of Neurology Washington University School of Medicine St Louis MO USA

E-mail leuthardtewudosiswustledu

Received 30 December 2010Accepted for publication 20 May 2011Published 10 June 2011Online at stacksioporgJNE8046013

AbstractElectrocorticography (ECoG)-based brainndashcomputer interface (BCI) systemshave emerged as a new signal platform for neuroprosthetic application ECoG-based platformshave shown significant promise for clinical application due to the high level of informationthat can be derived from the ECoG signal the signalrsquos stability and its intermediatenature of surgical invasiveness However before long-term BCI applications can be realizedit will be important to also understand how the cortical physiology alters with age Suchunderstanding may provide an appreciation for how this may affect the control signals utilizedby a chronic implant In this study we report on a large population of adult and pediatricinvasively monitored subjects to determine the impact that age will have on surface corticalphysiology We evaluated six frequency bandsmdashdelta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band (30ndash50 Hz) and high gamma band (76ndash100 Hz)mdashtoevaluate the effect of age on the magnitude of power change cortical area of activation andcortical networks When significant trends are evaluated as a whole it appears that the agingprocess appears to more substantively alter thalamocortical interactions leading to an increasein cortical inefficiency Despite this we find that higher gamma rhythms appear to be moreanatomically constrained with age while lower frequency rhythms appear to broaden in corticalinvolvement as time progresses From an independent signal standpoint this would favorhigh gamma rhythmsrsquo utilization as a separable signal that could be maintained chronically

(Some figures in this article are in colour only in the electronic version)

1 Introduction

Brainndashcomputer interface (BCI) research usingelectrocorticography (ECoG) as a signal substrate has6 Author to whom any correspondence should be addressed

received considerable attention as a potential clinical platformfor neuroprosthetic application Compared with scalp-recorded electroencephalographic (EEG) signals ECoG hasmuch larger signal magnitude increased spatial resolution(01 cm versus 50 cm for EEG) and a higher frequency

1741-256011046013+10$3300 1 copy 2011 IOP Publishing Ltd Printed in the UK

J Neural Eng 8 (2011) 046013 J Roland et al

bandwidth (0ndash500 versus 0ndash40 Hz for EEG) [4 5 12 46]These high frequencies known as gamma rhythms (gt 40 Hz)are especially notable because they appear to conveysignificant amount of information with regard to motorkinematics speech processing and reduced BCI trainingtimes [18 22 32 42 44 49] Gamma rhythms are thoughtto be produced by smaller cortical assemblies and showstrong correlations with neuronal action potential firings[16 37] Because ECoG electrodes do not penetrate the brainthey have been shown to have superior long-term stabilityin different animal models [6 8 23 25 51] In addition toits superior long-term stability a recent study showed thatthe neural substrate that encodes movements is also stableover many months [8] Taken together there is mountingevidence that ECoG may have important advantages for theBCI operation in the real world application

Currently the majority of studies of human ECoG requirethe use of invasively monitored human patients with intractableepilepsy Because of the clinical requirements of this scenariothe ability to study ECoG signals is constrained in regard tothe number of patients available and the duration in whichthese signals can be acquired (usually less than 2 weeks)While primate studies suggest that the ECoG signal is itselfstable over time [8] these findings do not indicate how theECoG physiology itself will change over time with agePrevious studies suggest that cortex and motor-associatedcortical signals are affected by age While it is known thatthe performance of motor tasks declines with age this isthought to be due to a multifactorial process related to bothcentral and peripheral causes [41] In the central nervoussystem there is evidence of cortical thinning thought to besecondary to alterations in white matter and corticalndashcorticalconnections [14] Additionally functional imaging and EEGdemonstrate wider areas of motor task-related brain activityin older subjects when compared to their younger populations[17 40] It has been posited that these changes are secondaryto a loss of cortical efficiency for performing a given motoroperation Due to age-related changes in the neurovasculatureand frequency limitations of EEG interpretation of theseresults is limited [36 39] As neuroprosthetic technologiesbegin to translate into chronic clinical implants how cortexchanges and how the associated signals change with agewill have important implications for the brain-derived devicecontrol

To assess the impact on how age affects surface corticalphysiology we examined ECoG recordings in a broad agerange of human subjects while performing a simple motortask We show in this study a differential effect of ageassociated with the frequency and anatomic distribution ofcortical activation Age appears to affect lower frequencyrhythms most greatly while gamma rhythms appear to bestable in anatomic distribution and amplitude modulationThese findings support the role of gamma band activations forthe stable and long-term use as a control signal for ECoG-BCIapplications

2 Methods

21 Data collection

Twenty-three patients with intractable focal epilepsy requiringinvasive monitoring with placement of subdural electrodearrays for seizure localization participated in this study(13 males 10 females) Their ages ranged from 11 to 59 yearsand they were from two institutions (Washington Universityin St Louis and University of Washington) The number ofelectrodes implanted ranged from 16 to 64 (mean 547 median64) All subjects had grid electrodes over sensorimotor cortex(11 right 12 left) and a seizure focus that did not involvesensorimotor cortex The number of electrodes located oversensorimotor cortex ranged from 1 to 13 (mean 69 median 7)Placement of all electrode arrays was determined by clinicalcriterion alone The ECoG electrode arrays (Ad-Tech RacineWI) consisted of 4 mm platinum electrodes (23 mm exposedsurface) with 1 cm interelectrode spacing (figure 1(A)) Thesampling frequency ranged from 300 to 1200 Hz and the signalwas band-pass filtered from 015 or 03 to 200 or 500 HzSampling and band-pass varied depending on an institutionalrecording system The study was approved independently bythe institutional review boards at the respective institutions andall participants gave informed consent

22 Electrode localization

Post-implant radiographs were used to identify electrodelocations (figure 1(B)) and derive normalized Talairachcoordinates using the Location on Cortex (LOC) program[28] Brodmann areas were then derived from thesecoordinates using a Talairach transformation database(httpwwwtalairachorgdaemonhtml) Electrode locationswere pooled for the young (age lt 25 years N = 10) andold (age gt 35 years N = 7) age groups The centroidand 95 confidence intervals for each group (figures 1(C)and (D)) were calculated and were not significantlydifferent The anteriorndashposterior and dorsalndashventralTalairach coordinate distributions among the age groupsoverlapped substantially (young dorsalndashventral mean =052 and sem = 12 anteriorndashposterior mean = 223 andsem = 10 old dorsalndashventral mean = 071 and sem =15 anteriorndashposterior mean = 227 and sem = 13)

23 Task

All cues for motor movement were delivered via a flat screenmonitor Visual cues were presented using the BCI2000software program In the context of brain mapping itsupports programmable presentation of audiovisual stimuliand simultaneous ECoG signal recordings [43] A cue wasgiven for the subject to perform a hand task consisting ofrepeatedly opening and closing the hand in a continuous andself-paced fashion The hand used was always the handcontralateral to the side where the grid electrodes were placedregardless of subject handedness Each movement period (atrial) was followed by an equal length inter-trial interval (ITIor rest period) The subject was instructed to perform the

2

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

Figure 1 ECoG placement and localization Intraoperative photograph illustrating the placement of a subdural ECoG electrode array (A)Post-operative lateral radiograph used for localizing ECoG electrodes (B) Cumulative distribution of electrode locations for the (C) young(lt25 years) and (D) old (gt35 years) age groups Centroid and 95 confidence intervals for the anteriorndashposterior and dorsalndashventralTalairach coordinates are represented by a green encircled cross-hair All electrodes were projected to the right hemisphere for illustration

Figure 2 Task diagram The experimental recording system was connected via splitter cable to provide ECoG input to the BCI2000research system simultaneous to continuous clinical monitoring The patient was instructed to open and close their hand contralateral to theside of grid implant regardless of subject handedness A second monitor driven by the BCI2000 software suite then provided visual cues forthe patient to perform the task during the trial period or wait calmly during the inter-trial interval (ITI) The task was self-paced opening andclosing of their hand continuously while cued with the word lsquohandrsquo displayed on the monitor

task only when the appropriate cue (the word lsquohandrsquo) waspresented and to stop moving and rest calmly when the cuedisappeared (blank screen) during the ITI period (figure 2)Due to clinical constraints motor metrics were not obtainablefrom all participants The trial and ITI periods were both 3 sin duration The number of trials ranged between 15 and 54depending on the patientrsquos ability to participate

24 Signal analysis

The power spectral density (PSD) was obtained from thelast 25 s of each trial in 1 Hz bins using a sliding 250 ms

Hanning window with 50 overlap Several frequency bandswere analyzed to examine the breadth of classically describedelectrophysiology as well as frequency bands of emergingimportance in the gamma range These bands includedthe delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta(13ndash30 Hz) low gamma band (30ndash50 Hz) and high gammaband (76ndash100 Hz) [9 21 35] For each bin the correlationcoefficient (r2) was calculated to detect a change in the spectralpower between rest and active conditions A t-statistic and thecorresponding p-value were calculated from the correlationcoefficient by t = r

radic(1 minus r2)(N minus 2) A p-value was

obtained for each bin and Bonferroni corrected for the number

3

J Neural Eng 8 (2011) 046013 J Roland et al

of electrodes A corrected p-value of less than 005 wasused to indicate a significant activation in that bin A binwith a significant p-value in the range of a previously definedfrequency band was used to indicate activation in the respectiveband The number of electrodes showing an activation for agiven band indicated the surface area of cortical activation

To determine the magnitude of amplitude modulation theelectrode with the highest correlation coefficient was chosenFor this electrode the average power during rest trials wassubtracted from the average power during movement trialsand the difference was divided by the average power duringrest trials to determine the percent change in power from thebaseline The bin within each of the respective bands with themaximum percent change in power for each individual wasthen selected for analysis The percentage of power change forthe respective frequency band accommodated the known 1f

dropoff of power and allowed for a normalized comparisonof power change across the frequency bands where lowerfrequency bands have a higher power and higher frequencybands have a lower power

To measure changes in cortical networks the mutualinformation (MI) statistic between band-pass filtered ECoGtime series was calculated Mutual information is aquantification of the information gained about one randomvariable X from the measurement of a second variable Y [45] IfX and Y are independent then their mutual information is zeroHowever if knowing the value of Y reduces the uncertainty ofthe value of X then the mutual information between the twosystems is in the normalized range of (0 1]

The average amount of information gained from anymeasurement of X is the entropy (H) of X The mutualinformation between X and Y is defined as MI(XY) = H(X) ndashH(X|Y) where H(X|Y) is the information about the value of Xgained by a measurement of Y

Formally the entropy H(X) is defined as

H(X) =int infin

minusinfinPX(x) log 2(PX(x))

where PX(x) is the probability that X = x in the system XWhile it is usually not feasible to exactly define probabilitiesfor real systems they can be estimated as histograms Thusin practice when there are discrete time-series measurementsX = x1 x2 xn and Y = y1 y2 yn that take onmeasurement values i1 i2 im and j 1 j 2 jmrespectively the mutual information between systems X and Ycan be estimated as

M(X Y ) =j=1msumi=1m

sumt=1n

(xt = i cap yt = j) log2

times(sum

t=1n

(xt = i cap yt = j)

((sumt=1n

xt = i

)

times(sum

t=1n

yt = j

)))

Mutual information is related to covariance but it has theadvantage that the relationship between the values of X and Ydoes not need to be linear Any relationship that is consistentover the time that the measurements are taken will increase

the mutual information score These relationships detected bymutual information were interpreted as an interaction betweenthe cortical sites or common modulation of cortical sites byanother source

This was then used to define a network of corticalcommunication Each raw time series was first band-passfiltered for the respective bands of interest The MI wascalculated between all electrode pairs for each band-limitedtime series over a range of time lags The time lags werecreated by shifting one of the two time series from minus150 to+150 ms by one time sample at a time For each electrode pairthe time lag with the maximum MI score was selected for eachtrial and ITI period Significant differences in the average MIbetween the trial and ITI periods were determined by Studentrsquost-test A p-value of less than 005 after Bonferroni correctionfor the number of electrodes was used to indicate a change inthe network with the task

For each frequency band the degree of correlationwith age for these three statistics (amplitude modulationarea of activation and changing networks) was determinedby Pearsonrsquos correlation coefficient A p-value for eachcorrelation coefficient was calculated as described aboveand a p-value of less than 005 was used to determinestatistical significance Outliers were identified as samplepoints greater than three standard deviations from the meanand were discarded prior to calculating correlation Subjectsthat showed no electrodes with significant power modulationarea of activation or network changes within the band beinganalyzed were not included in the respective analysis In thepower modulation analysis this resulted in N = 7 for delta N =13 for theta N = 19 for alpha N = 21 for beta N = 16 for lowgamma and N = 19 for high gamma band showing significantchanges In the area of activation analysis this resulted in N =7 for delta N = 13 for theta N = 18 for alpha N = 21 for betaN = 15 for low gamma and N = 19 for high gamma bandIn network changes analysis this resulted in N = 14 for deltaN = 19 for theta N = 19 for alpha N = 20 for beta N = 15for low gamma and N = 19 for high gamma band

3 Results

31 Amplitude modulation as a function of age

The fundamental electrophysiologic feature underlying ECoGand EEG-based BCI applications relies on detecting changesin the spectral power as it is modulated by the subjectTherefore we first compared the magnitude of power changeacross the age and frequency band For each frequencyband the magnitude of spectral power modulated duringtask performance was plotted against age with polynomialtrendlines (figures 3(A)ndash(F)) Those that were statisticallysignificant were plotted as solid lines and those that did notreach statistical significance were plotted as dashed linesGenerally the magnitude of percent change in power forall low-frequency bands (delta theta alpha and beta) wasnot large The delta band however did show a significantcorrelation with age (R2 = 06 and p = 004) (figure 3(A))The remaining low-frequency bands did not have a significant

4

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 3 Magnitude of power modulation For each frequency band the magnitude of spectral power modulated during task performance isplotted against age with polynomial trendlines These bands included the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz)low gamma band (30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005level while the dashed trendlines indicate correlation coefficients that did not reach statistical significance

relationship to age (theta R2 = 016 and p = 017 alphaR2 = 011 and p = 016 beta R2 = 010 and p =016) (figures 3(B)ndash(D)) The high-frequency bands showedsignificant correlations with age (low gamma R2 = 033and p = 0021 high gamma R2 = 0236 and p = 0035)(figures 3(E) and (F)) These associations are best representedby an inverse parabola with a nadir near the mean subject ageof 30 years The low gamma band trend of power modulationis appreciably more flat relative to the high gamma band

32 Area of cortical activation as a function of age

The area of cortex activated with task performance was thenexamined The region of activated cortex (as defined bystatistically significant power modulation p lt 005) wasrepresented by the number of electrodes among a regularlyspaced grid that showed active signal changes For eachfrequency band the area of activation was plotted against agewith polynomial trendlines (figures 4(A)ndash(F)) Those that werestatistically significant were plotted as solid lines and thosethat did not reach statistical significance were plotted as dashedlines The alpha and beta bands had significant increasesin area of cortex activated with age (alpha R2 = 025 andp = 0027 beta R2 = 036 and p = 0004) (figures 5(C) and(D)) The region of cortical activation increased approximately10ndash15 cm2 for both bands The remaining low-frequencybands did not show statistically significant correlations withage (delta R2 = 051 and p = 024 theta R2 = 036 andp = 022) (figures 4(A) and (B)) The low gamma band alsodemonstrated a statistically significant association with age(R2 = 06 and p = 0006) Although significant the trend

was smaller ranging from 1 to a maximum of 3 electrodesof cortical activation (figure 4(E)) The high gamma band didnot show a significant correlation with area activated by age(R2 = 027 and p = 006) (figure 4(F))

33 Cortical networks as a function of age

Changes in cortical networks with task performance arerepresented by a number of electrode pairs showing asignificant (p lt 005) change in the mutual information scoreFor each frequency band the number of electrode pairs wasplotted against age with polynomial trendlines (figures 5(A)ndash(F)) Those that were statistically significant were plotted assolid lines those that did not reach statistical significancewere plotted as dashed lines The beta band showed asignificant increase associated with age (R2 = 06 and p =00001) (figure 5(D)) All other bands did not show significantassociations with age (delta R2 = 018 and p = 013 thetaR2 = 00004 and p = 094 alpha R2 = 017 and p = 049low gamma R2 = 023 and p = 04 high gamma R2 = 034and p = 016) (figures 5(A)ndash(C) (E) and (F))

4 Discussion

This is the first time that invasive electrocorticography hasbeen applied to study the effect of age on the motor corticalfunction The analysis of high-quality invasively acquiredhuman data provides important insight into the effects of ageon the motor cortical function that cannot be extrapolatedfrom studies using non-invasive modalities (eg MRI MEG

5

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 4 Cortical area of activation The area of cortex activated with task performance is represented by the number of electrodes among aregularly spaced grid showing active signal changes For each frequency band the area of activation is plotted against age with polynomialtrendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band (30ndash50 Hz) andhigh gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while the dashed trendlinesindicate correlation coefficients that did not reach statistical significance

or EEG) or animal models alone This study demonstratesseveral trends that occur across multiple frequency bandswhich are likely indicative of underlying alterations (or lackthereof) in their respective neural sources Both the stableand dynamic character of these physiologic signals havean important bearing on their use for BCI operation whenconsidered as control features that will be used through thecourse of the patientrsquos life The more stable and independentthe physiology through the aging process the more likelythat signal will remain a reliable control feature for BCIoperation When significant trends are evaluated as a wholeit appears that the aging process appears to more substantivelyalter thalamocortical interactions leading to an increase incortical inefficiency Despite this we find that higher gammarhythms appear to be more anatomically constrained with age

while lower frequency rhythms appear to broaden in corticalinvolvement as time progresses From an independent signalstandpoint this would favor high gamma rhythmsrsquo utilizationas a separable signal that could be maintained chronically

Early investigations into cortical changes with agewere conducted by cytological and imaging studies [33]Cytological methods necessitated ex vivo specimens butallowed direct visualization of cortical tissue at the histologicallevel Initial methodological complications led to debateconcerning the etiology of cortical thinning that is associatedwith increasing age While initial reports suggested thatneuronal populations decrease more recent evidence suggeststhat neuronal cell counts do not significantly change [31 34]but rather stromal or white matter changes may be implicatedin the observed cortical thinning [33] With the advent of more

6

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 5 Changes in cortical networks Changes in cortical networks with task performance are represented by the number of electrodepairs showing a significant change in mutual information score For each frequency band the number of electrode pairs is plotted against agewith polynomial trendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band(30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while thedashed trendlines indicate correlation coefficients that did not reach statistical significance

advanced imaging techniques similar studies were conductedin vivo with nuclear magnetic resonance imaging (MRI) [30]These techniques quantified the amount of white and graymatter in volunteer research participants The results of thesestructural imaging studies tend to concur by attributing theeffect of cortical thinning in normal aging to white matterchanges [11 13] While this type of data provided importantinsight into healthy and pathological aging it was not ableto provide any information about the active physiologicalprocesses of the brain

Histological studies were further complemented withthe emergence of functional MRI (fMRI) and non-invasiveelectrophysiology [38] fMRI relies on blood-oxygen-leveldependent (BOLD) changes in the MR signal and is thoughtto be a measure of the hemodynamic response to local cortical

activity [3] In the context of aging several research paradigmshave reported a wider area of BOLD response to simple motoractivity in the contralateral cerebral hemisphere as well as anincreased bilateral representation associated with increasingage [17 26] Consistent with these findings EEG studiesyield similar results to those of functional imaging suggestinga wider cortical area of activation and increased bilateralrepresentation associated with increasing age [40] Both thesefindings have been posited to represent a loss of corticalefficiency with age [7] Namely broader cortical resourcesare required to perform the same motor task However boththese approaches suffer from technical limitations that prohibitdefinitive conclusions regarding this process The alteration ofcerebrovascular autoregulation with age makes interpretationof BOLD changes more challenging as to whether these

7

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

of electrodes A corrected p-value of less than 005 wasused to indicate a significant activation in that bin A binwith a significant p-value in the range of a previously definedfrequency band was used to indicate activation in the respectiveband The number of electrodes showing an activation for agiven band indicated the surface area of cortical activation

To determine the magnitude of amplitude modulation theelectrode with the highest correlation coefficient was chosenFor this electrode the average power during rest trials wassubtracted from the average power during movement trialsand the difference was divided by the average power duringrest trials to determine the percent change in power from thebaseline The bin within each of the respective bands with themaximum percent change in power for each individual wasthen selected for analysis The percentage of power change forthe respective frequency band accommodated the known 1f

dropoff of power and allowed for a normalized comparisonof power change across the frequency bands where lowerfrequency bands have a higher power and higher frequencybands have a lower power

To measure changes in cortical networks the mutualinformation (MI) statistic between band-pass filtered ECoGtime series was calculated Mutual information is aquantification of the information gained about one randomvariable X from the measurement of a second variable Y [45] IfX and Y are independent then their mutual information is zeroHowever if knowing the value of Y reduces the uncertainty ofthe value of X then the mutual information between the twosystems is in the normalized range of (0 1]

The average amount of information gained from anymeasurement of X is the entropy (H) of X The mutualinformation between X and Y is defined as MI(XY) = H(X) ndashH(X|Y) where H(X|Y) is the information about the value of Xgained by a measurement of Y

Formally the entropy H(X) is defined as

H(X) =int infin

minusinfinPX(x) log 2(PX(x))

where PX(x) is the probability that X = x in the system XWhile it is usually not feasible to exactly define probabilitiesfor real systems they can be estimated as histograms Thusin practice when there are discrete time-series measurementsX = x1 x2 xn and Y = y1 y2 yn that take onmeasurement values i1 i2 im and j 1 j 2 jmrespectively the mutual information between systems X and Ycan be estimated as

M(X Y ) =j=1msumi=1m

sumt=1n

(xt = i cap yt = j) log2

times(sum

t=1n

(xt = i cap yt = j)

((sumt=1n

xt = i

)

times(sum

t=1n

yt = j

)))

Mutual information is related to covariance but it has theadvantage that the relationship between the values of X and Ydoes not need to be linear Any relationship that is consistentover the time that the measurements are taken will increase

the mutual information score These relationships detected bymutual information were interpreted as an interaction betweenthe cortical sites or common modulation of cortical sites byanother source

This was then used to define a network of corticalcommunication Each raw time series was first band-passfiltered for the respective bands of interest The MI wascalculated between all electrode pairs for each band-limitedtime series over a range of time lags The time lags werecreated by shifting one of the two time series from minus150 to+150 ms by one time sample at a time For each electrode pairthe time lag with the maximum MI score was selected for eachtrial and ITI period Significant differences in the average MIbetween the trial and ITI periods were determined by Studentrsquost-test A p-value of less than 005 after Bonferroni correctionfor the number of electrodes was used to indicate a change inthe network with the task

For each frequency band the degree of correlationwith age for these three statistics (amplitude modulationarea of activation and changing networks) was determinedby Pearsonrsquos correlation coefficient A p-value for eachcorrelation coefficient was calculated as described aboveand a p-value of less than 005 was used to determinestatistical significance Outliers were identified as samplepoints greater than three standard deviations from the meanand were discarded prior to calculating correlation Subjectsthat showed no electrodes with significant power modulationarea of activation or network changes within the band beinganalyzed were not included in the respective analysis In thepower modulation analysis this resulted in N = 7 for delta N =13 for theta N = 19 for alpha N = 21 for beta N = 16 for lowgamma and N = 19 for high gamma band showing significantchanges In the area of activation analysis this resulted in N =7 for delta N = 13 for theta N = 18 for alpha N = 21 for betaN = 15 for low gamma and N = 19 for high gamma bandIn network changes analysis this resulted in N = 14 for deltaN = 19 for theta N = 19 for alpha N = 20 for beta N = 15for low gamma and N = 19 for high gamma band

3 Results

31 Amplitude modulation as a function of age

The fundamental electrophysiologic feature underlying ECoGand EEG-based BCI applications relies on detecting changesin the spectral power as it is modulated by the subjectTherefore we first compared the magnitude of power changeacross the age and frequency band For each frequencyband the magnitude of spectral power modulated duringtask performance was plotted against age with polynomialtrendlines (figures 3(A)ndash(F)) Those that were statisticallysignificant were plotted as solid lines and those that did notreach statistical significance were plotted as dashed linesGenerally the magnitude of percent change in power forall low-frequency bands (delta theta alpha and beta) wasnot large The delta band however did show a significantcorrelation with age (R2 = 06 and p = 004) (figure 3(A))The remaining low-frequency bands did not have a significant

4

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 3 Magnitude of power modulation For each frequency band the magnitude of spectral power modulated during task performance isplotted against age with polynomial trendlines These bands included the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz)low gamma band (30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005level while the dashed trendlines indicate correlation coefficients that did not reach statistical significance

relationship to age (theta R2 = 016 and p = 017 alphaR2 = 011 and p = 016 beta R2 = 010 and p =016) (figures 3(B)ndash(D)) The high-frequency bands showedsignificant correlations with age (low gamma R2 = 033and p = 0021 high gamma R2 = 0236 and p = 0035)(figures 3(E) and (F)) These associations are best representedby an inverse parabola with a nadir near the mean subject ageof 30 years The low gamma band trend of power modulationis appreciably more flat relative to the high gamma band

32 Area of cortical activation as a function of age

The area of cortex activated with task performance was thenexamined The region of activated cortex (as defined bystatistically significant power modulation p lt 005) wasrepresented by the number of electrodes among a regularlyspaced grid that showed active signal changes For eachfrequency band the area of activation was plotted against agewith polynomial trendlines (figures 4(A)ndash(F)) Those that werestatistically significant were plotted as solid lines and thosethat did not reach statistical significance were plotted as dashedlines The alpha and beta bands had significant increasesin area of cortex activated with age (alpha R2 = 025 andp = 0027 beta R2 = 036 and p = 0004) (figures 5(C) and(D)) The region of cortical activation increased approximately10ndash15 cm2 for both bands The remaining low-frequencybands did not show statistically significant correlations withage (delta R2 = 051 and p = 024 theta R2 = 036 andp = 022) (figures 4(A) and (B)) The low gamma band alsodemonstrated a statistically significant association with age(R2 = 06 and p = 0006) Although significant the trend

was smaller ranging from 1 to a maximum of 3 electrodesof cortical activation (figure 4(E)) The high gamma band didnot show a significant correlation with area activated by age(R2 = 027 and p = 006) (figure 4(F))

33 Cortical networks as a function of age

Changes in cortical networks with task performance arerepresented by a number of electrode pairs showing asignificant (p lt 005) change in the mutual information scoreFor each frequency band the number of electrode pairs wasplotted against age with polynomial trendlines (figures 5(A)ndash(F)) Those that were statistically significant were plotted assolid lines those that did not reach statistical significancewere plotted as dashed lines The beta band showed asignificant increase associated with age (R2 = 06 and p =00001) (figure 5(D)) All other bands did not show significantassociations with age (delta R2 = 018 and p = 013 thetaR2 = 00004 and p = 094 alpha R2 = 017 and p = 049low gamma R2 = 023 and p = 04 high gamma R2 = 034and p = 016) (figures 5(A)ndash(C) (E) and (F))

4 Discussion

This is the first time that invasive electrocorticography hasbeen applied to study the effect of age on the motor corticalfunction The analysis of high-quality invasively acquiredhuman data provides important insight into the effects of ageon the motor cortical function that cannot be extrapolatedfrom studies using non-invasive modalities (eg MRI MEG

5

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 4 Cortical area of activation The area of cortex activated with task performance is represented by the number of electrodes among aregularly spaced grid showing active signal changes For each frequency band the area of activation is plotted against age with polynomialtrendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band (30ndash50 Hz) andhigh gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while the dashed trendlinesindicate correlation coefficients that did not reach statistical significance

or EEG) or animal models alone This study demonstratesseveral trends that occur across multiple frequency bandswhich are likely indicative of underlying alterations (or lackthereof) in their respective neural sources Both the stableand dynamic character of these physiologic signals havean important bearing on their use for BCI operation whenconsidered as control features that will be used through thecourse of the patientrsquos life The more stable and independentthe physiology through the aging process the more likelythat signal will remain a reliable control feature for BCIoperation When significant trends are evaluated as a wholeit appears that the aging process appears to more substantivelyalter thalamocortical interactions leading to an increase incortical inefficiency Despite this we find that higher gammarhythms appear to be more anatomically constrained with age

while lower frequency rhythms appear to broaden in corticalinvolvement as time progresses From an independent signalstandpoint this would favor high gamma rhythmsrsquo utilizationas a separable signal that could be maintained chronically

Early investigations into cortical changes with agewere conducted by cytological and imaging studies [33]Cytological methods necessitated ex vivo specimens butallowed direct visualization of cortical tissue at the histologicallevel Initial methodological complications led to debateconcerning the etiology of cortical thinning that is associatedwith increasing age While initial reports suggested thatneuronal populations decrease more recent evidence suggeststhat neuronal cell counts do not significantly change [31 34]but rather stromal or white matter changes may be implicatedin the observed cortical thinning [33] With the advent of more

6

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 5 Changes in cortical networks Changes in cortical networks with task performance are represented by the number of electrodepairs showing a significant change in mutual information score For each frequency band the number of electrode pairs is plotted against agewith polynomial trendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band(30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while thedashed trendlines indicate correlation coefficients that did not reach statistical significance

advanced imaging techniques similar studies were conductedin vivo with nuclear magnetic resonance imaging (MRI) [30]These techniques quantified the amount of white and graymatter in volunteer research participants The results of thesestructural imaging studies tend to concur by attributing theeffect of cortical thinning in normal aging to white matterchanges [11 13] While this type of data provided importantinsight into healthy and pathological aging it was not ableto provide any information about the active physiologicalprocesses of the brain

Histological studies were further complemented withthe emergence of functional MRI (fMRI) and non-invasiveelectrophysiology [38] fMRI relies on blood-oxygen-leveldependent (BOLD) changes in the MR signal and is thoughtto be a measure of the hemodynamic response to local cortical

activity [3] In the context of aging several research paradigmshave reported a wider area of BOLD response to simple motoractivity in the contralateral cerebral hemisphere as well as anincreased bilateral representation associated with increasingage [17 26] Consistent with these findings EEG studiesyield similar results to those of functional imaging suggestinga wider cortical area of activation and increased bilateralrepresentation associated with increasing age [40] Both thesefindings have been posited to represent a loss of corticalefficiency with age [7] Namely broader cortical resourcesare required to perform the same motor task However boththese approaches suffer from technical limitations that prohibitdefinitive conclusions regarding this process The alteration ofcerebrovascular autoregulation with age makes interpretationof BOLD changes more challenging as to whether these

7

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 3 Magnitude of power modulation For each frequency band the magnitude of spectral power modulated during task performance isplotted against age with polynomial trendlines These bands included the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz)low gamma band (30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005level while the dashed trendlines indicate correlation coefficients that did not reach statistical significance

relationship to age (theta R2 = 016 and p = 017 alphaR2 = 011 and p = 016 beta R2 = 010 and p =016) (figures 3(B)ndash(D)) The high-frequency bands showedsignificant correlations with age (low gamma R2 = 033and p = 0021 high gamma R2 = 0236 and p = 0035)(figures 3(E) and (F)) These associations are best representedby an inverse parabola with a nadir near the mean subject ageof 30 years The low gamma band trend of power modulationis appreciably more flat relative to the high gamma band

32 Area of cortical activation as a function of age

The area of cortex activated with task performance was thenexamined The region of activated cortex (as defined bystatistically significant power modulation p lt 005) wasrepresented by the number of electrodes among a regularlyspaced grid that showed active signal changes For eachfrequency band the area of activation was plotted against agewith polynomial trendlines (figures 4(A)ndash(F)) Those that werestatistically significant were plotted as solid lines and thosethat did not reach statistical significance were plotted as dashedlines The alpha and beta bands had significant increasesin area of cortex activated with age (alpha R2 = 025 andp = 0027 beta R2 = 036 and p = 0004) (figures 5(C) and(D)) The region of cortical activation increased approximately10ndash15 cm2 for both bands The remaining low-frequencybands did not show statistically significant correlations withage (delta R2 = 051 and p = 024 theta R2 = 036 andp = 022) (figures 4(A) and (B)) The low gamma band alsodemonstrated a statistically significant association with age(R2 = 06 and p = 0006) Although significant the trend

was smaller ranging from 1 to a maximum of 3 electrodesof cortical activation (figure 4(E)) The high gamma band didnot show a significant correlation with area activated by age(R2 = 027 and p = 006) (figure 4(F))

33 Cortical networks as a function of age

Changes in cortical networks with task performance arerepresented by a number of electrode pairs showing asignificant (p lt 005) change in the mutual information scoreFor each frequency band the number of electrode pairs wasplotted against age with polynomial trendlines (figures 5(A)ndash(F)) Those that were statistically significant were plotted assolid lines those that did not reach statistical significancewere plotted as dashed lines The beta band showed asignificant increase associated with age (R2 = 06 and p =00001) (figure 5(D)) All other bands did not show significantassociations with age (delta R2 = 018 and p = 013 thetaR2 = 00004 and p = 094 alpha R2 = 017 and p = 049low gamma R2 = 023 and p = 04 high gamma R2 = 034and p = 016) (figures 5(A)ndash(C) (E) and (F))

4 Discussion

This is the first time that invasive electrocorticography hasbeen applied to study the effect of age on the motor corticalfunction The analysis of high-quality invasively acquiredhuman data provides important insight into the effects of ageon the motor cortical function that cannot be extrapolatedfrom studies using non-invasive modalities (eg MRI MEG

5

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 4 Cortical area of activation The area of cortex activated with task performance is represented by the number of electrodes among aregularly spaced grid showing active signal changes For each frequency band the area of activation is plotted against age with polynomialtrendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band (30ndash50 Hz) andhigh gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while the dashed trendlinesindicate correlation coefficients that did not reach statistical significance

or EEG) or animal models alone This study demonstratesseveral trends that occur across multiple frequency bandswhich are likely indicative of underlying alterations (or lackthereof) in their respective neural sources Both the stableand dynamic character of these physiologic signals havean important bearing on their use for BCI operation whenconsidered as control features that will be used through thecourse of the patientrsquos life The more stable and independentthe physiology through the aging process the more likelythat signal will remain a reliable control feature for BCIoperation When significant trends are evaluated as a wholeit appears that the aging process appears to more substantivelyalter thalamocortical interactions leading to an increase incortical inefficiency Despite this we find that higher gammarhythms appear to be more anatomically constrained with age

while lower frequency rhythms appear to broaden in corticalinvolvement as time progresses From an independent signalstandpoint this would favor high gamma rhythmsrsquo utilizationas a separable signal that could be maintained chronically

Early investigations into cortical changes with agewere conducted by cytological and imaging studies [33]Cytological methods necessitated ex vivo specimens butallowed direct visualization of cortical tissue at the histologicallevel Initial methodological complications led to debateconcerning the etiology of cortical thinning that is associatedwith increasing age While initial reports suggested thatneuronal populations decrease more recent evidence suggeststhat neuronal cell counts do not significantly change [31 34]but rather stromal or white matter changes may be implicatedin the observed cortical thinning [33] With the advent of more

6

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 5 Changes in cortical networks Changes in cortical networks with task performance are represented by the number of electrodepairs showing a significant change in mutual information score For each frequency band the number of electrode pairs is plotted against agewith polynomial trendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band(30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while thedashed trendlines indicate correlation coefficients that did not reach statistical significance

advanced imaging techniques similar studies were conductedin vivo with nuclear magnetic resonance imaging (MRI) [30]These techniques quantified the amount of white and graymatter in volunteer research participants The results of thesestructural imaging studies tend to concur by attributing theeffect of cortical thinning in normal aging to white matterchanges [11 13] While this type of data provided importantinsight into healthy and pathological aging it was not ableto provide any information about the active physiologicalprocesses of the brain

Histological studies were further complemented withthe emergence of functional MRI (fMRI) and non-invasiveelectrophysiology [38] fMRI relies on blood-oxygen-leveldependent (BOLD) changes in the MR signal and is thoughtto be a measure of the hemodynamic response to local cortical

activity [3] In the context of aging several research paradigmshave reported a wider area of BOLD response to simple motoractivity in the contralateral cerebral hemisphere as well as anincreased bilateral representation associated with increasingage [17 26] Consistent with these findings EEG studiesyield similar results to those of functional imaging suggestinga wider cortical area of activation and increased bilateralrepresentation associated with increasing age [40] Both thesefindings have been posited to represent a loss of corticalefficiency with age [7] Namely broader cortical resourcesare required to perform the same motor task However boththese approaches suffer from technical limitations that prohibitdefinitive conclusions regarding this process The alteration ofcerebrovascular autoregulation with age makes interpretationof BOLD changes more challenging as to whether these

7

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 4 Cortical area of activation The area of cortex activated with task performance is represented by the number of electrodes among aregularly spaced grid showing active signal changes For each frequency band the area of activation is plotted against age with polynomialtrendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band (30ndash50 Hz) andhigh gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while the dashed trendlinesindicate correlation coefficients that did not reach statistical significance

or EEG) or animal models alone This study demonstratesseveral trends that occur across multiple frequency bandswhich are likely indicative of underlying alterations (or lackthereof) in their respective neural sources Both the stableand dynamic character of these physiologic signals havean important bearing on their use for BCI operation whenconsidered as control features that will be used through thecourse of the patientrsquos life The more stable and independentthe physiology through the aging process the more likelythat signal will remain a reliable control feature for BCIoperation When significant trends are evaluated as a wholeit appears that the aging process appears to more substantivelyalter thalamocortical interactions leading to an increase incortical inefficiency Despite this we find that higher gammarhythms appear to be more anatomically constrained with age

while lower frequency rhythms appear to broaden in corticalinvolvement as time progresses From an independent signalstandpoint this would favor high gamma rhythmsrsquo utilizationas a separable signal that could be maintained chronically

Early investigations into cortical changes with agewere conducted by cytological and imaging studies [33]Cytological methods necessitated ex vivo specimens butallowed direct visualization of cortical tissue at the histologicallevel Initial methodological complications led to debateconcerning the etiology of cortical thinning that is associatedwith increasing age While initial reports suggested thatneuronal populations decrease more recent evidence suggeststhat neuronal cell counts do not significantly change [31 34]but rather stromal or white matter changes may be implicatedin the observed cortical thinning [33] With the advent of more

6

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 5 Changes in cortical networks Changes in cortical networks with task performance are represented by the number of electrodepairs showing a significant change in mutual information score For each frequency band the number of electrode pairs is plotted against agewith polynomial trendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band(30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while thedashed trendlines indicate correlation coefficients that did not reach statistical significance

advanced imaging techniques similar studies were conductedin vivo with nuclear magnetic resonance imaging (MRI) [30]These techniques quantified the amount of white and graymatter in volunteer research participants The results of thesestructural imaging studies tend to concur by attributing theeffect of cortical thinning in normal aging to white matterchanges [11 13] While this type of data provided importantinsight into healthy and pathological aging it was not ableto provide any information about the active physiologicalprocesses of the brain

Histological studies were further complemented withthe emergence of functional MRI (fMRI) and non-invasiveelectrophysiology [38] fMRI relies on blood-oxygen-leveldependent (BOLD) changes in the MR signal and is thoughtto be a measure of the hemodynamic response to local cortical

activity [3] In the context of aging several research paradigmshave reported a wider area of BOLD response to simple motoractivity in the contralateral cerebral hemisphere as well as anincreased bilateral representation associated with increasingage [17 26] Consistent with these findings EEG studiesyield similar results to those of functional imaging suggestinga wider cortical area of activation and increased bilateralrepresentation associated with increasing age [40] Both thesefindings have been posited to represent a loss of corticalefficiency with age [7] Namely broader cortical resourcesare required to perform the same motor task However boththese approaches suffer from technical limitations that prohibitdefinitive conclusions regarding this process The alteration ofcerebrovascular autoregulation with age makes interpretationof BOLD changes more challenging as to whether these

7

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

(A) (B)

(C) (D)

(E) (F)

Figure 5 Changes in cortical networks Changes in cortical networks with task performance are represented by the number of electrodepairs showing a significant change in mutual information score For each frequency band the number of electrode pairs is plotted against agewith polynomial trendlines These bands include the delta (lt4 Hz) theta (4ndash8 Hz) alpha (8ndash13 Hz) beta (13ndash30 Hz) low gamma band(30ndash50 Hz) and high gamma band (76ndash100 Hz) [9 21 35] The solid lines indicate statistical significance at the p = 005 level while thedashed trendlines indicate correlation coefficients that did not reach statistical significance

advanced imaging techniques similar studies were conductedin vivo with nuclear magnetic resonance imaging (MRI) [30]These techniques quantified the amount of white and graymatter in volunteer research participants The results of thesestructural imaging studies tend to concur by attributing theeffect of cortical thinning in normal aging to white matterchanges [11 13] While this type of data provided importantinsight into healthy and pathological aging it was not ableto provide any information about the active physiologicalprocesses of the brain

Histological studies were further complemented withthe emergence of functional MRI (fMRI) and non-invasiveelectrophysiology [38] fMRI relies on blood-oxygen-leveldependent (BOLD) changes in the MR signal and is thoughtto be a measure of the hemodynamic response to local cortical

activity [3] In the context of aging several research paradigmshave reported a wider area of BOLD response to simple motoractivity in the contralateral cerebral hemisphere as well as anincreased bilateral representation associated with increasingage [17 26] Consistent with these findings EEG studiesyield similar results to those of functional imaging suggestinga wider cortical area of activation and increased bilateralrepresentation associated with increasing age [40] Both thesefindings have been posited to represent a loss of corticalefficiency with age [7] Namely broader cortical resourcesare required to perform the same motor task However boththese approaches suffer from technical limitations that prohibitdefinitive conclusions regarding this process The alteration ofcerebrovascular autoregulation with age makes interpretationof BOLD changes more challenging as to whether these

7

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

changes are representative of an actual change in corticalphysiology or an effect of altered neurovascular coupling[2 10] While EEG does not have this limitation it suffersfrom a lower spatial resolution smaller frequency bandwidthand a limitation of signal quality This lower spatial resolutionis due to the averaging of potentials from an area of cortexproportional to the distance between the recording electrodeand the cortical source [12] Therefore EEG signals representthe mean of multiple cortical ensembles and may not bestrepresent the local cortical physiology

The use of ECoG provides a signal that gives anintegrated perspective in which reasonable spatial resolutionand excellent temporal and spectral resolution are combinedThe findings in this study are both consistent with andsupplement previously published findings on the age-relatedchanges in cortex Taken together the findings in this worksupport the loss of neural efficiency in the brainrsquos ability toperform a motor task This is supported in many levels ofanalysis First the alteration in the magnitude of power is seenin several important frequency bands When gamma powerchanges were analyzed there was a u-shaped function notedGiven that higher gamma frequency power changes are closelyassociated action potential spiking these findings would beconsistent with previously published studies that infer learningand optimal cortical utilization (ie lowest amount of cortexneeded to perform the task) to occur into early adulthood[15 24 37 47 52] The fact that high gamma power changesare lowest in their early thirties would imply that cortex ismost efficient at that age Additionally in the delta rhythmthere is a modest linear decline in the magnitude of powerwith age Most notably in anesthesia and sleep studies thisvery low frequency rhythm has been associated with thalamicprocessing [1] We would posit that this represents a decliningthalamic input that occurs with time Second when the corticalarea of activation is considered we find that there is a robustexpansion of cortical area (several centimeters to 10ndash15 cm2)that demonstrates alpha and beta rhythm power modulationin association with a motor task which is consistent withprevious EEG studies [40] These lower frequency bandsare thought to represent post-synaptic potentials secondary tothalamocortical projections When the gamma rhythms wereexamined there was a modest statistically significant increasein areas of activation (increase in 1ndash2 cm2) associated withthe lower gamma frequency band (the rhythm most closelyassociated with gabaergic interneurons [48]) but not with thehigher gamma band (the rhythm most closely correlated withsingle unit firing [15 24 37]) These anatomic findings wouldargue that there is a greater degree of cortex modulated bythe thalamus but that the actual region of cortex engagedand activated in the task remains unchanged This notionof altered thalamic modulation is further supported by amutual information analysis We found evidence of increasedmutual information among cortical sites primarily in the betarange that likely represents a common modulatory source (egthe thalamus) We would interpret this increased sharedinformation with age to be interpreted as a degenerativeprocess whereby the thalamus is modulating a larger areaof cortex in order to achieve the same task Taken together

rather than the decline in brain function being purely due toloss of cortical efficiency we would posit that the inefficiencyin neural processing is more due to a decline in thalamicprocessing with less specificity in cortical modulation

These findings have an important bearing on ECoG andEEG-based neuroprosthetics in the future When consideringa signal platform the more stable the underlying physiologythe more likely the control will be reliable with time Namelythe signal that a user is employing for control should not besubjected to drift due to changes in the underlying cellularsubstrate Additionally it will be important that if multiplesignals are being used for device control those control featuresremain independent with time Taking these parameters intoconsideration the findings in this study support the use of highgamma rhythms as being the optimal signal substrate for long-term ECoG-BCI usage As mentioned previously these higherfrequency rhythms above 75 Hz are more closely correlatedwith action potential firing [15 24 37] and their stabilityover time in regards to anatomic distribution of activationduring a task is also consistent with cytological and structuralimaging studies that report the stability of neuronal cell countsand gray matter with age respectively [11 13 30 31 34]Although there were indeed changes in the magnitude of powerchange with age we would posit that the anatomic stabilityand lack of shared mutual information would favor that thesehigher frequency signals would remain independent with timeOf note the lower gamma rhythms (30ndash50 Hz) howevershowed a modest several centimeter increase in anatomic areaengaged with age Although small this is still significant whenconsidering that an ECoG implant may utilize microarraysin the future and the recording electrodes would be spacedmillimeters apart [19 49] Consistent with previous EEGfindings [40] the lower frequency rhythms showed the mostsubstantial changes with age in terms of anatomic distributionof activation (alpha and beta) as well as notable increasesin shared mutual information (beta) This has relevance toboth EEG and ECoG-based BCI The significant increase inthe area of activation and mutual information is concerningbecause with time these signals could lose independence (ifmultiple signals are being used [27 50]) and thus compromisethe ongoing control capability of the BCI

Although this is a unique study which utilizes a highnumber of invasively monitored patients across a large agerange there are some important limitations and caveats worthnoting Most notably all the patients that participated in thisstudy have chronic epilepsy Therefore these data must beinterpreted with some degree of caution in the setting of thepotentially pathologically altered brain structure or functionIt is possible that the changes identified across age could be dueto the effects of chronic epilepsy versus age itself Althoughpossible we would posit this to be less likely given thatour low-frequency findings parallel closely with those seenin previous EEG studies [40] in normal healthy volunteersAdditionally this patient population by definition has partialepilepsy which is localized to a discrete region which makesthem suitable candidates for invasive monitoring If a personrsquosseizures appear to originate from bilateral hemispheres orare otherwise non-focal in origin then a surgical treatment

8

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

is not appropriate Thus by excluding electrodes over theidentified seizure focus from our research data the potentialadverse effects of the disease and its impact on these findingsare probably limited Also of note recordings for this studyutilized macroelectrode recordings (1 cm spacing) which wereindicated for the clinical need of seizure localization Givenrecent studies using microscale cortical recordings for BCIcontrol [20] the likely form factor for a BCI implant in thefuture will likely use microarrays of the order of millimeterspacing Thus how age will affect microscale corticalnetworks will need to be further defined Finally it shouldbe noted that we were assessing cortical changes that occurduring the performance of a simple overt motor task Whensignals are used directly for BCI operation this can alter thenascent physiology such that it can be distinct from the corticalactivations during the screening task that defined the controlfeatures originally [29] Thus how the chronic use of a BCIimpacts these signals is currently untested and may also playa role in the evolution of these signals over time

In summary these findings are significant because theyaugment previous assertions that the brain has reducedefficiency of its resources with age This study points to aless selective thalamocortical recruitment and corticocorticalinformation processing with the aging brain This distinctionpoints to a larger role of thalamic projections rather thanintrinsic cortical activation to explain age-dependent changesin cortical engagement with a motor task Taken togetherusing higher frequency gamma rhythms which most closelyreflect local cortical ensembles may best support chronic BCIapplications in the future

Acknowledgments

This work was funded by the Doris Duke Foundation theChildrenrsquos Discovery Institute and the McDonnell Center forHigher Brain Function We wish to acknowledge Dr JeffreyOjemann for contributing patient data We would also liketo thank the patients who participated Their cooperationand diligence makes this science possible and is greatlyappreciated

References

[1] Alkire M Haier R and Fallon J 2000 Toward a unified theoryof narcosis brain imaging evidence for a thalamocorticalswitch as the neurophysiologic basis of anesthetic-inducedunconsciousness Conscious Cogn 9 370ndash86

[2] Ances B M Liang C L Leontiev O Perthen J E Fleisher A SLansing A E and Buxton R B 2009 Effects of aging oncerebral blood flow oxygen metabolism and bloodoxygenation level dependent responses to visual stimulationHum Brain Mapp 30 1120ndash32

[3] Attwell D and Iadecola C 2002 The neural basis of functionalbrain imaging signals Trends Neurosci 25 621ndash5

[4] Ball T Kern M Mutschler I Aertsen A and Schulze-BonhageA 2009 Signal quality of simultaneously recorded invasiveand non-invasive EEG Neuroimage 46 708ndash16

[5] Boulton A A Baker G B and Vanderwolf C H 1990Neurophysiological Techniques II Applications to NeuralSystems (Clifton NJ Humana Press)

[6] Bullara L A Agnew W F Yuen T G Jacques S andPudenz R H 1979 Evaluation of electrode array material forneural prostheses Neurosurgery 5 681ndash6

[7] Cabeza R Anderson N D Locantore J K and McIntosh A R2002 Aging gracefully compensatory brain activity inhigh-performing older adults Neuroimage 17 1394ndash402

[8] Chao Z C Nagasaka Y and Fujii N 2010 Long-termasynchronous decoding of arm motion usingelectrocorticographic signals in monkeys FrontNeuroeng 3 3

[9] Crone N E Miglioretti D L Gordon B and Lesser R P 1998Functional mapping of human sensorimotor cortex withelectrocorticographic spectral analysis II Event-relatedsynchronization in the gamma band Brain121 Pt 12 2301ndash15

[10] DrsquoEsposito M Zarahn E Aguirre G K and Rypma B 1999 Theeffect of normal aging on the coupling of neural activity tothe bold hemodynamic response Neuroimage 10 6ndash14

[11] Fotenos A F Snyder A Z Girton L E Morris J C andBuckner R L 2005 Normative estimates of cross-sectionaland longitudinal brain volume decline in aging and ADNeurology 64 1032ndash9

[12] Freeman W J Holmes M D Burke B C and Vanhatalo S 2003Spatial spectra of scalp EEG and EMG from awake humansClin Neurophysiol 114 1053ndash68

[13] Guttmann C R G Jolesz F A Kikinis R Killiany R JMoss M B Sandor T and Albert M S 1998 White matterchanges with normal aging Neurology 50 972ndash8

[14] Haug H and Eggers R 1991 Morphometry of the human cortexcerebri and corpus striatum during aging Neurobiol Aging12 336ndash8

[15] Heldman D A Wang W Chan S S and Moran D W 2004Local field potential spectral tuning in primary motor cortexSoc Neurosci Abstr 42122

[16] Heldman D A Wang W Chan S S and Moran D W 2006Local field potential spectral tuning in motor cortex duringreaching IEEE Trans Neural Syst Rehabil Eng 14 180ndash3

[17] Hutchinson S Kobayashi M Horkan C M Pascual-Leone AAlexander M P and Schlaug G 2002 Age-related differencesin movement representation Neuroimage 17 1720ndash8

[18] Kellis S Miller K Thomson K Brown R House P andGreger B Decoding spoken words using local fieldpotentials recorded from the cortical surface J Neural Eng7 056007

[19] Leuthardt E C Freudenberg Z Bundy D and Roland J 2009Microscale recording from human motor corteximplications for minimally invasive electrocorticographicbrainndashcomputer interfaces Neurosurg Focus 27 E10

[20] Leuthardt E C Gaona C M Sharma M Szarma N Roland JFreudenberg Z Solis J Breshears J and Schalk G 2011Using the electrocorticographic speech network to control abrainndashcomputer interface in humans J Neural Eng 8 036004

[21] Leuthardt E C Miller K J Anderson N Schalk G Dowling JMoran D Miller J and Ojemann J G 2007Electrocorticographic frequency alteration mapping aclinical technique for mapping motor cortex Neurosurgery60 260ndash70

[22] Leuthardt E C Schalk G Wolpaw J R Ojemann J Gand Moran D W 2004 A brain-computer interface usingelectrocorticographic signals in humans J Neural Eng1 63ndash71

[23] Loeb G E Walker A E Uematsu S and Konigsmark B W 1977Histological reaction to various conductive and dielectricfilms chronically implanted in the subdural spaceJ Biomed Mater Res 11 195ndash210

[24] Manning J R Jacobs J Fried I and Kahana M J 2009Broadband shifts in local field potential power spectra arecorrelated with single-neuron spiking in humansJ Neurosci 29 13613ndash20

9

J Neural Eng 8 (2011) 046013 J Roland et al

[25] Margalit E et al 2003 Visual and electrical evoked responserecorded from subdural electrodes implanted above thevisual cortex in normal dogs under two methods ofanesthesia J Neurosci Methods 123 129ndash37

[26] Mattay V S Fera F Tessitore A Hariri A R Das SCallicott J H and Weinberger D R 2002 Neurophysiologicalcorrelates of age-related changes in human motor functionNeurology 58 630ndash5

[27] McFarland D J Sarnacki W A and Wolpaw J RElectroencephalographic (EEG) control ofthree-dimensional movement J Neural Eng7 036007

[28] Miller K J Makeig S Hebb A O Rao R P denNijs Mand Ojemann J G 2007 Cortical electrode localization fromX-rays and simple mapping for electrocorticographicresearch the lsquoLocation on Cortexrsquo (LOC) package forMATLAB J Neurosci Methods 162 303ndash8

[29] Miller K J Schalk G Fetz E E den Nijs M Ojemann J Gand Rao R P 2010 Cortical activity during motor executionmotor imagery and imagery-based online feedback ProcNatl Acad Sci USA 107 4430ndash5

[30] Minati L Grisoli M and Bruzzone M G 2007 MRspectroscopy functional MRI and diffusion-tensor imagingin the aging brain a conceptual review J GeriatrPsychiatry Neurol 20 3ndash21

[31] Morrison J H and Hof P R 1997 Life and death of neurons inthe aging brain Science 278 412ndash9

[32] Pei X Leuthardt E C Gaona C M Brunner P Wolpaw J Rand Schalk G 2011 Spatiotemporal dynamics ofelectrocorticographic high gamma activity during overt andcovert word repetition Neuroimage 54 2960ndash72

[33] Peters A 2002 The effects of normal aging on myelin andnerve fibers a review J Neurocytol 31 581ndash93

[34] Peters A Morrison J Rosene D and Hyman B 1998 Featurearticle are neurons lost from the primate cerebral cortexduring normal aging Cereb Cortex 8 295ndash300

[35] Pfurtscheller G 1999 Event-Related Desynchronization (ERD)and Event Related Synchronization (ERS) (Baltimore MDWilliams and Wilkins)

[36] Pfurtscheller G and Cooper R 1975 Frequency dependence ofthe transmission of the EEG from cortex to scalpElectroencephalogr Clin Neurophysiol 38 93ndash6

[37] Ray S Crone N E Niebur E Franaszczuk P J and Hsiao S S2008 Neural correlates of high-gamma oscillations (60ndash200Hz) in macaque local field potentials and their potentialimplications in electrocorticography J Neurosci28 11526ndash36

[38] Reuter-Lorenz P A 2002 New visions of the aging mind andbrain Trends Cogn Sci 6 394ndash400

[39] Riecker A Grodd W Klose U Schulz J B Groschel KErb M Ackermann H and Kastrup A 2003 Relation

between regional functional MRI activation and vascularreactivity to carbon dioxide during normal aging J CerebBlood Flow Metab 23 565ndash73

[40] Sailer A Dichgans J and Gerloff C 2000 The influence ofnormal aging on the cortical processing of a simple motortask Neurology 55 979ndash85

[41] Sallis J F 2000 Age-related decline in physical activity asynthesis of human and animal studies Med Sci SportsExerc 32 1598ndash600

[42] Schalk G Kubanek J Miller K J Anderson N RLeuthardt E C Ojemann J G Limbrick D Moran DGerhardt L A and Wolpaw J R 2007 Decodingtwo-dimensional movement trajectories usingelectrocorticographic signals in humans J Neural Eng4 264ndash75

[43] Schalk G McFarland D J Hinterberger T Birbaumer Nand Wolpaw J R 2004 BCI2000 a general-purposebrainndashcomputer interface (BCI) system IEEE TransBiomed Eng 51 1034ndash43

[44] Schalk G Miller K J Anderson N R Wilson J A Smyth M DOjemann J G Moran D W Wolpaw J R and Leuthardt E C2008 Two-dimensional movement control usingelectrocorticographic signals in humans J Neural Eng5 75ndash84

[45] Shannon C E and Weaver W 1949 The Mathematical Theoryof Communication (Champaign IL University of IllinoisPress)

[46] Slutzky M W Jordan L R Krieg T Chen M Mogul D Jand Miller L E Optimal spacing of surface electrode arraysfor brain-machine interface applications J Neural Eng7 26004

[47] Smits-Engelsman B C Van Galen G P and Duysens J 2004Force levels in uni- and bimanual isometric tasks affectvariability measures differently throughout lifespan MotorControl 8 437ndash49

[48] Sukov W and Barth D S 1998 Three-dimensional analysis ofspontaneous and thalamically evoked gamma oscillations inauditory cortex J Neurophysiol 79 2875ndash84

[49] Wang W et al 2009 Human motor cortical activity recordedwith Micro-ECoG electrodes during individual fingermovements Conf Proc IEEE Eng Med Biol Soc2009 586ndash9

[50] Wolpaw J R and McFarland D J 2004 Control of atwo-dimensional movement signal by a noninvasivebrain-computer interface in humans Proc Natl Acad SciUSA 101 17849ndash54

[51] Yuen T G Agnew W F and Bullara L A 1987 Tissue responseto potential neuroprosthetic materials implanted subdurallyBiomaterials 8 138ndash41

[52] Zanone P G 1990 Tracking with and without target in 6- to15-year-old boys J Motor Behav 22 225ndash49

10