Psychiatry Research: Neuroimaging Section 90 Ž1999. 125]140 Relationship between brain electrical activity and cortical perfusion in normal subjects Andrew F. Leuchter a,b,c,U , Sebastian H.J. Uijtdehaage b, Ian A. Cook a,b, Ruth O’Harae, Mark Mandelkern d,f a Quantitati¨ e EEG Laboratory, UCLA Neuropsychiatric Institute and Hospital, 760 Westwood Plaza, Los Angeles, CA 90024, USA b Department of Psychiatry and Biobeha¨ ioral Sciences, UCLA School of Medicine, Los Angeles, CA, USA c Medication De¨ elopment Research Unit, West Los Angeles VA Medical Center, Los Angeles, CA, USA d Department of Nuclear Medicine, West Los Angeles VA Medical Center, Los Angeles, CA, USA e Department of Psychiatry and Beha¨ ioral Sciences, Stanford Uni¨ ersity, Stanford, CA, USA f Department of Physics, Uni¨ ersity of California, Ir¨ ine, CA, USA Received 30 December 1998; received in revised form 25 January 1999; accepted 31 January 1999 Abstract Cerebral glucose uptake and perfusion are accepted as tightly coupled measures of energy utilization in both normal and diseased brain. The coupling of brain electrical activity to perfusion has been demonstrated, however, only in the presence of chronic brain disease. Very few studies have examined the relationship between cerebral electrical activity and energy utilization in normal brain tissue. To clarify this relationship, we performed 33 Ž . H 15 2 O-positron emission tomography PET scans in six normal subjects both at rest and during a simple motor task, and acquired surface-recorded quantitative electroencephalogram ŽQEEG. data simultaneously with isotope injection. We examined the associations between cerebral perfusion directly underlying each recording electrode and three QEEG measures Žabsolute power, relative power, and cordance.. All EEG measures had moderately strong coupling with perfusion at most frequency bands, although the directions of the associations differed from those previously reported in subjects with stroke or dementia. Of the three QEEG measures examined, cordance had the strongest relationship with perfusion Žmultiple R 2 s 0.58.. Cordance and PET were equally effective in detecting lateralized activation associated with the motor task, while EEG power did not detect this activation. Electrodes in the concordant state had a significantly higher mean perfusion than those in the discordant state. These results U Corresponding author. Tel.: q1-310-825-0207; fax: q1-310-825-7642; e-mail: email@example.com 0925-4927r99r$ - see front matter Q 1999 Elsevier Science Ireland Ltd. All rights reserved. PII: S 0 9 2 5 - 4 9 2 7 Ž 9 9 . 0 0 0 0 6 - 2 126 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 indicate that normal brain electrical activity has a moderately strong association with cerebral perfusion. Cordance may be the most useful QEEG measure for monitoring cerebral perfusion in subjects without chronic brain disease. Q 1999 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Cerebral perfusion; Quantitative electroencephalogram ŽQEEG.; Positron emission tomography ŽPET.; Cordance; Normal subjects 1. Introduction Cerebral glucose uptake and blood flow long ago were hypothesized to be comparable measures of energy utilization ŽRoy and Sherrington, 1890.. This hypothesis now has been tested with imaging techniques, such as autoradiography, positron emission tomography ŽPET., and single photon emission computed tomography ŽSPECT.. In normal subjects, cerebral glucose uptake and blood flow generally are accepted as tightly coupled measures of cerebral energy utilization ŽDes Rossiers et al., 1974; Sokoloff, 1977, 1981.. Brain electrical activity represents the single greatest demand on cerebral metabolism ŽErecinska and Silver, 1989., suggesting that measurement of electrical energy also should be coupled to cerebral metabolism and perfusion. Berger Ž1938. first hypothesized that the rhythmic activity in the surface-recorded electroencephalogram ŽEEG. closely reflected brain metabolic activity. Interestingly, however, most previous studies have shown that EEG reflects cerebral energy utilization accurately only under conditions of extreme dysfunction. Animal models using blood vessel occlusion ŽCartheuser, 1988. or metabolic suppression with medication ŽKlementavicius et al., 1996. have demonstrated strong associations between the cerebral metabolic rate for oxygen and EEG power and frequency. Similarly, studies of human subjects suffering from stroke ŽTolonen and Sulg, 1981; Nagata et al., 1982; Nagata, 1988., degenerative brain diseases ŽStigsby et al., 1981; Wszolek et al., 1992; Passero et al., 1995; Valladeres-Neto et al., 1995., or epilepsy ŽJibiki et al., 1994. have found that cerebral perfusion and metabolism have a negative association with slow-wave energy and a positive association with alpha energy. Very few studies have examined associations between electrical activity and cerebral perfusion in the normal brain. Most studies asserted to examine normal brain actually focused upon the undamaged cerebral hemisphere in stroke patients ŽMelamed et al., 1975; Nagata, 1989; Nagata et al., 1989.. It now is known, however, that the contralateral hemisphere in stroke patients may show changes in metabolism and perfusion ŽSerrati et al., 1994., perhaps reflecting transcallosal fiber degeneration ŽIglesias et al., 1996.. Some studies utilized elderly volunteers with incompletely characterized health status ŽObrist et al., 1963. or patient volunteers ŽIngvar and Risberg, 1967; Ingvar et al., 1976; Ingvar, 1979. who suffered from chronic psychiatric illnesses andror conditions that are currently recognized as risk factors for brain disease Ži.e. alcohol abuse, atherosclerosis .. These studies reported relationships between cerebral electrical activity and perfusion which ranged from weak ŽObrist et al., 1963. to moderately strong ŽIngvar and Risberg, 1967; Ingvar et al., 1976; Ingvar, 1979., but it is not clear that the subjects examined were truly representative of normal function. Only two studies have examined the association between EEG power and PET scanning Žusing the fluorodeoxyglucose-18 technique. ŽBuchsbaum et al., 1984. or SPECT scanning Žusing the Xenon-133 technique. ŽOkyere et al., 1986. obtained simultaneously in normal subjects. Although both groups found moderately strong associations between electrical activity Žin the alpha band. and perfusion, consistent associations were limited to the occipital regions. Examination of other brain regions showed a variable relationship between EEG power and metabolism, with both positive and negative associations in the same EEG frequency band in different brain regions ŽBuchsbaum et al., 1984.. A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 Because of inconsistencies in the methods and results from previous studies, the relationship between surface-recorded EEG in different frequency bands and the perfusion of underlying brain tissue remains unclear. We performed the current study to clarify the associations between quantitative EEG ŽQEEG. measures and cerebral perfusion Žusing 15 O-positron emission tomography. in normal subjects at rest and while performing a motor activation task. A secondary aim of this study was to derive a QEEG index predicting relative perfusion. We examined three QEEG measures: absolute power, relative power, and cordance. Cordance integrates absolute and relative power, and in previous work in subjects with brain disease Ži.e. stroke, dementia. has shown more robust and consistent associations with cerebral perfusion Žas measured by HMPAO-SPECT. than either power measure alone ŽLeuchter et al., 1994a,b.. 2. Materials and methods 2.1. Subjects Six right-handed male subjects Žages 20]30, mean age 28. with no history of psychiatric, medical, or neurologic illness were recruited from the community. All subjects were assessed with a clinical history, neurologic examination, and magnetic resonance imaging ŽMRI. scanning to confirm the absence of neurologic disease. All experiments were approved by the UCLA Human Subjects Protection Committee, and subjects’ consent was obtained according to the Declaration of Helsinki. 2.2. Acti¨ ation task The subjects were examined in a darkened, sound-attenuated room, in both the resting condition and while they were performing a simple motor task Žsqueezing a foam rubber ball with the right or left hand.. Both the resting and motor task conditions were repeated in the eyes-open and eyes-closed states. For the resting condition, subjects were maintained in the maximally alert state for 2 min through frequent re-alerting. For 127 the motor task condition, subjects were instructed to squeeze a foam rubber ball with either the right or left hand continuously for a period of 2]3 min Žincluding a pre-recording period of 1 min.. Subjects underwent scanning in the following order of conditions: Ž1. resting, eyes closed; Ž2. right hand squeezing, eyes open; Ž3. left hand squeezing, eyes open; Ž4. right hand squeezing, eyes closed; Ž5. resting, eyes open; Ž6. right hand squeezing, eyes closed; Ž7. right hand squeezing, eyes open; and Ž8. left hand squeezing, eyes open. 2.3. PET scanning procedure PET scans were performed using a Siemens ECAT 953 scanner which has an axial acceptance of 10.5 cm and generates 31 transaxial planes. The patient was positioned on the gantry so that the scan planes were parallel to the cantho]meatal line. PET data were acquired starting immediately after isotope injection in 30-s frames, and data from the second frame Žstarting 30 s after injection, representing tracer uptake in the brain. were used for processing. Fig. 1 summarizes the recording procedure during the motor task. Exactly 25 mCi of H 2 15 O-tracer were injected intravenously as a bolus while the subject was in the resting state or Žfor the activation conditions. 60 s after the subject began performing the motor task. Subjects were permitted to rest for 10 min between motor task sessions, with eight scans performed per subject. We performed 48 H 15 2 OPET scans, but because of a failure to obtain simultaneous EEG data, excessive EEG artifacts during data acquisition, or technical difficulties with the PET scanner, 15 scans were eliminated from the analysis for a total of 33 scans. 2.4. QEEG recording QEEG data were acquired simultaneously with PET scanning using a QND electroencephalograph ŽNeurodata; Pasadena, CA., with a sampling rate of 256 pointsrchannelrsecond, high frequency filter of 70 Hz., and low frequency filter of 0.5 Hz. Electrodes were placed according to an enhanced montage based upon the International 128 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 Fig. 1. Sequence and time course of the PETrEEGrtask protocol. 10]20 System ŽFig. 2. using a custom electrode cap ŽElectroCap Inc; Eaton, OH.. EEG data were selected from the time frame 15]60 s after injection. The EEG data sampling window started before the PET data sampling window, since cerebral blood flow changes may lag behind electrical activity by as much as 10 s ŽToga et al., 1995.. The EEG data sampling window was slightly longer than that for PET, to facilitate selection of artifact-free data. Any data that were contaminated by drowsiness, muscle, motion, or other artifacts were deleted. All artifact-free EEG data recorded in the designated time frame were selected for further processing; the amount varied between 12 and 30 s per EEG. Absolute EEG power Žmeasured in mV 2 . was calculated in a series of overlapping 4-Hz bands between 0 and 40 Hz Ž0]4 Hz, 2]6 Hz, 4]8 Hz, etc.. using a fast Fourier transform ŽPress et al., 1986.. Power was calculated according to the single-source Žlocal Laplacian. technique described by Hjorth Ž1975., in which voltage values from surrounding nearest-neighbor electrodes are averaged to obtain voltage attributable to a single electrode. This montage was selected since the local Laplacian derivation more accurately reflects the electrical activity attributable to a given electrode than common-reference EEG data ŽGevins, 1990.. 2.5. Cordance algorithm Cordance combines absolute and relative QEEG power measurements, characterizing both the magnitude of and the relationship between Fig. 2. Map showing 36 electrodes used for EEG recording. This montage is based upon the International 10]20 System, with additional electrodes placed along diagonals between electrode sites to provide enhanced coverage of the frontal and parietal association cortices. Lines indicate ‘nearest neighbor’ electrode pairs used for EEG power and cordance calculations. Boxes over the central area delineate those electrodes over the left and right motor strips which were used to detect lateralized activation. absolute and relative power at each recording electrode. The revised cordance algorithm reported here is based upon three steps in the processing of power, each of which is designed to refine the association between power and cerebral perfusion. These steps are: Ž1. reattribution of power from bipolar pairs of electrodes to individual electrodes; Ž2. spatial normalization of absolute and relative power across brain areas; and Ž3. characterization of the association between normalized absolute and relative power measures. Each step in processing of power, as well as the rationale for each step, is discussed separately below. 2.5.1. Reattributing power There is no single electrode reference which is ideal for calculating power values. All referencing methods may obscure or distort information in the EEG signal, by underestimating power measurements from electrodes close to the reference andror contributing local signal effects from the A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 reference to distant electrodes. One indicator of the capacity of a reference to reflect accurately the activity originating under a recording electrode is to examine the association between power calculated using different referencing methods and the perfusion underlying each electrode. Cook et al. Ž1998. recently compared linked-ears, reattributed amplitude Ži.e. single-source or Hjorth transformation., and reattributed power referencing methods in terms of the strength of association between power at a recording electrode and perfusion underlying that electrode Žusing H 15 2 Opositron emission tomography.. This study found that the reattributed power measure showed the strongest association with perfusion underlying the recording electrode. To calculate reattributed power, absolute power was calculated for each bipolar pair of nearestneighbor electrodes shown in the grid of electrodes in Fig. 2 Ževery electrode paired with its nearest neighbors.. Absolute power was reattributed from electrode pairs to individual electrodes, by averaging EEG power for all electrode pairs which included that electrode ŽCook et al., 1998.. Before averaging, any channels containing significant muscle or other artifacts were deleted from the grid. If the deletion of channels for reason of artifact left any electrode with fewer than two remaining nearest neighbors, that electrode was deleted completely. After averaging, a square-root transformation was performed to minimize skewness and kurtosis ŽLeuchter et al., 1993.. Lastly, relative EEG power Žabsolute power in a frequency band divided by total power for the entire spectrum. was calculated based upon the reattributed absolute power. 2.5.2. Spatial normalization of power Absolute power is affected by factors that may not directly be related to brain function Ži.e. interelectrode distances, characteristics of tissue between electrode and brain. and is difficult to compare between individuals since total power varies widely between individuals. Absolute power therefore commonly is normalized with respect to the energy across the frequency spectrum Ži.e. relative power, or the percentage of energy in a frequency band.. While relative power controls 129 for major sources of power variation between individuals, it also may suppress information which be contained in the absolute power measure. Leuchter et al. Ž1993. reported that absolute and relative power measures were complementary measures of brain activity; while they are strongly interrelated, they also contain unique information regarding brain function. In order to retain information from both absolute and relative power, they each were normalized independently across electrode sites Žspatial normalization. using z-scores. Mean absolute and relative power was calculated across all electrode sites in each frequency band f. Z-scores then were calculated for each electrode site s in each frequency band. This process yielded normalized absolute and relative power values for each electrode site w A norm Ž s, f . and R norm Ž s, f . , respectivelyx. 2.5.3. Relating normalized power measures It previously has been reported that absolute and relative power can be affected differently by brain dysfunction. Either absolute or relative power may be little changed by brain dysfunction; conversely, absolute power in a band over an area of dysfunction may decrease while relative power in that same band may increase ŽLeuchter et al., 1994a.. In order to retain the information content of both absolute and relative power, A norm and R norm first are summed to yield a single value that indicates the cordance value w ZŽ s, f . x at each electrode site: ZŽ s, f . s A normŽ s, f . q R normŽ s, f . Next, A norm and R norm are examined to determine if they are above or below their respective mean power values. One of four conditions exists at each electrode site in each frequency band: Ž1. A norm and R norm are both above their respective means; Ž2. A norm is below but R norm is above their respective means; Ž3. A norm and R norm both are below their respective means; or Ž4. A norm is above but R norm is below their respective mean values for that frequency band. Each of these conditions corresponds to one of the quadrants in Fig. 3. If an electrode is characterized by 130 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 Fig. 3. Diagram showing the methods by which cordance relates normalized absolute and relative power. Quadrant in the above figure is indicated both by polar coordinates Ž08, 908, 1808, 2708. and by quadrant numbers ŽI, II, III, IV.. After normalization of the absolute and relative power values described in the methods, the normalized values w A norm Ž s, f . and R norm Ž s, f . x are summed to create the cordance value w ZŽ s, f . x. In addition to a cordance value, each electrode also has a categorical value indicating whether A norm Ž s, f . and R norm Ž s, f . both are above or below their respective means Žthe concordant state, Fig. 3a., or whether one is above while the other is below its respective mean Žthe discordant state, Fig. 3b.. Z values both of which are above or below the mean values Ži.e. quadrant I or III, or polar coordinates between 0 and 908 or 180 and 2708, respectively., that electrode shows concordance Ži.e. absolute and relative power are positively associated; see Fig. 3a.. If conversely either A norm or R norm is above while the other is below their respective mean values Ži.e. quadrant II or IV, or polar coordinates between 90 and 1808 or 270 and 3608, respectively., that electrode shows discordance Ži.e. absolute and relative power are negatively associated; see Fig. 3b.. Each electrode site in each frequency band therefore has a cordance value ZŽ s, f . , as well as a categorical descriptor indicating whether it is in the concordant or discordant state. 2.6. Integrating QEEG data and PET images The technique for measuring cerebral perfusion under each electrode has been described in detail elsewhere ŽCook et al., 1998. and is described only briefly here. This technique allowed for accurate and consistent spatial registration of each EEG electrode with the immediately under- lying cerebral cortex. Prior to PET scanning, Lucite markers containing 22 NaCl were placed over each EEG electrode recording site to permit visualization of the sites on PET images ŽFig. 3.. Images were processed on a Tatung microCOMPstation5 SPARC-compatible workstation ŽTatung Science and Technology, Milpitas, CA. running AVS software ŽAdvanced Visualization System, Waltham, MA.. With the AVS software, a threedimensional PET image was reconstructed based upon the summation of the axial image data across slices. The centers of the 22 NaCl fiducial markers were identified, and lines were automatically drawn from each electrode location to a center point in the brain; this point is defined as the mid-point in the x- and y-directions in the axial plane best showing the basal ganglia. Each line served as the central axis for a cylindrical volume of cerebral cortex under each electrode site. The outer boundary of this volume was a 2.5-cm2 circular region of interest ŽROI. centered on the line segment, placed at the external level of the cortex Žwhere the pixel count exceeded background levels.; 2.5 cm2 is the cortical surface area A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 which has been shown to generate the unique activity recorded at a surface EEG electrode ŽGevins, 1990.. The inner boundary of the cylinder was the corresponding circle at a depth of 1.0 cm below the cortical surface. These 2.5-cm2 = 1.0-cm cylindrical volumes therefore spanned the cortical ribbon and constituted a digital sample of cortex underlying each electrode. The actual PET counts within these regions were summated to determine the perfusion value for each electrode site. These automatically computed volumes obviated the need for hand drawing of ROIs and the inaccuracies that would arise from estimating the cross-sections as these cylindrical ROIs traversed sequential image slices. To mitigate the slight fluctuations of the 25 mCi H 15 2 O tracers, PET values were scaled from 0 Žthe lowest count. to 1 Žthe highest count.. The resulting PET values therefore reflected relative perfusion. 2.7. Association between QEEG measures and perfusion Each combined EEG-PET recording resulted in an array of paired observations. Each pair consisted of an estimate of relative perfusion and a QEEG measure. For each 4 Hz band, relative perfusion was regressed separately on EEG power and ZŽ s, f . . The number of electrodes available for analysis in each individual scan was utilized as a weight in the regression equation. Each regression analysis was based on a total of 670 observations. Residuals were systematically examined, and no outliers Ždefined as standardized residuals ) 2. or other violations of the regression model were observed. Data from all scans were pooled and relative perfusion was regressed on dummycoded subject and condition variables to account for subject and condition effects. In addition, the mean relative perfusion values for electrodes showing concordance and discordance were calculated separately, and differences in mean perfusion were examined using t-tests. 2.8. Deri¨ ation of a QEEG measure of perfusion A tentative QEEG measure of relative perfusion was derived by determining the optimal com- 131 bination of frequency bands predicting PET values. In addition, the accuracy with which perfusion and the EEG measures each detected activation during the motor tasks was assessed. Two regions of interest were defined over the motor strip bilaterally, defined by the following EEG electrodes Žleft: T3, C3, FC5, FC1; right: T4, C4, FC6, FC2. Žregions are indicated by the rectangular boxes in Fig. 2.. Corresponding PET regions underlying these electrodes also were identified. For each individual motor activation session, we determined whether PET, cordance, and EEG power accurately detected lateralized activation Ždefined as a maximum activation value in the region of interest contralateral to the hand squeezing.. The proportion of tests showing accurate lateralization for each measure was calculated and examined using a binomial test to determine if accuracy in lateralization exceeded chance levels. 3. Results 3.1. Associations between QEEG and perfusion The strength of the associations between power or cordance and relative perfusion is displayed in Fig. 4, where the magnitude of the partial correlation coefficients is graphed as a function of frequency. In most frequency bands, power and cordance w ZŽ s, f . x showed significant associations with relative perfusion. For all EEG measures the relationship with perfusion was triphasic: positive associations were seen in the 4-Hz bands which had a lower bound below 6 Hz; negative associations were seen in the bands which had a lower bound between 6 and 10 Hz; and positive associations were seen in the bands which had a lower bound at or above 12 Hz Žexcept for absolute power, which had a negligible association between 12 and 18 Hz.. Among the EEG measures, the strongest associations with perfusion were obtained with ZŽ s, f . . The associations between ZŽ s, f . and relative perfusion were similar in all conditions ŽFig. 5.. Only the resting eyes-open condition yielded a deviant pattern of partial correlations, none of which were significant. Importantly, from all EEGs in 132 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 Fig. 4. Plot showing the partial correlation coefficient between EEG power and cordance values w ZŽ s, f . x, and relative perfusion as a function of frequency band. Statistical significance is indicated by horizontal lines representing the magnitude at which a correlation coefficient attains significance: dashed line Ž- - - -. for Ps 0.05; dotted line Ž? ? ?. for Ps 0.01; dotted]dashed line Ž ] ? ] . for Ps 0.001. this condition more than 12 electrodes were removed due to artifacts. As we established in a later stage of the analyses Žsee below., the validity of ZŽ s, f . became questionable if more than 12 electrodes were removed. A stepwise regression analysis was used to find the optimal combination of frequency bands to predict relative perfusion. ZŽ s, f . values for the respective 4-Hz band were offered as predictors of perfusion Žadjusted for subject and condition effects .. The results revealed that ZŽ s, f . for the 8]12-Hz Ž‘alpha’. band and for the 20]24-Hz band proved to be relatively strong predictors of perfusion wmultiple R s 0.51, F Ž11, 658. s 21.07 P- .001x. The association between the cordance measures Ž8]12 and 20]24 Hz. and perfusion was further enhanced if the scans for the eyes closed conditions were analyzed separately from those for the eyes-open condition weyes closed: R s 0.58, F Ž8, 295. s 16.70, P- 0.0001; eyes open: R s 0.49, F Ž8, 357. s 14.74, P- 0.0001, respectivelyx. We performed additional analyses examining the associations between perfusion and power for individual electrodes, and for groups of up to 8 electrodes clustered on a regional basis. None of the correlations were significant Ž r F 0.15.. This finding probably reflected the limitations of this dataset. First, there was a limited number of subjects. Second, there was a limited range of relative perfusion values in these normal subjects. Third, some electrodes had to be deleted from the montage in each recording because of problems with muscle artifact. An analysis of these data on an electrode-by-electrode, or even region-by-region basis, further limited the number of data points and diminished the range of the PET values in any one analysis. Electrodes were grouped by whether they were A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 133 Fig. 5. Plot showing the partial correlation coefficient between Z and relative perfusion for each condition Žresting state vs. hand squeezing, eyes open vs. eyes closed.. Key: RH ec s right hand squeezing, eyes closed; RH eo s right hand squeezing, eyes open; Rest eo s resting state, eyes open; Rest ec s resting state, eyes closed; Left eo s left hand squeezing, eyes open. concordant or discordant in the 20]24-Hz frequency band Ž n s 484., which had the strongest association with perfusion. The concordant electrodes had significantly higher mean relative perfusion than the discordant electrodes Ž0.58 vs. 0.42. Ž t s 3.21, P- .001.. Eyes closed: 3.2. Deri¨ ation of formulae to predict perfusion Eyes open: Relative Perfusion s 0.49516 y0.0472 w Z Ž 8]12Hz.x q0.04025 w Z Ž 20]24Hz.x Relative Perfusion s 0.504821 Regression equations were calculated for each individual scan with ZŽ s, f . for the 8]12- and 20]24-Hz bands as predictors. An optimized regression formula was derived by calculating the weighted mean of the intercepts and of the regression coefficients with the number of electrodes in each scan as weights. The eyes-open and eyes-closed conditions were analyzed separately and yielded the following preliminary equations: y0.02309 w Z Ž 8]12Hz.x q0.032559w Z Ž 20]24Hz.x where ZŽ8]12 Hz. and ZŽ20]24 Hz. are ZŽ s, f . for the 8]12-Hz band and the 20]24-Hz band, respectively. For each individual scan, we estimated perfusion of all available electrode sites using the formulae Žsome electrodes were deleted from the 134 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 montage due to artifacts .. Thus, we obtained pairwise values of estimated perfusion and ‘real’ perfusion as rendered by the PET. The correlation coefficient between those two sets of values was the validity coefficient. We found that if more than 12 electrodes were deleted from the 36-electrode montage, the validity coefficient dropped sharply for many recordings Ž- 0.2.. We therefore recalculated the perfusion equations holding out EEG recordings from which more than 12 electrodes were removed. The resulting equations were: Eyes closed: Relative Perfusion s 0.50406 y0.04836 w Z Ž 8]12Hz.x q0.03913 w Z Ž 20]24Hz.x Fig. 6. Distribution of the validity coefficients of the cordance-estimated perfusion values. A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 Table 1 Accuracy of detecting lateralized activation Ždefined as a maximum activation value in the region of interest contralateral to hand squeezing. Misses P Žbinomial distribution. Measure Hits PET Cordance QEEG power Ž8]12 Hz. QEEG power Ž20]24 Hz. 15 16 9 8 0.07 0.05 11 13 0.15 12 12 0.16 Eyes open: Relative Perfusion s 0.48500 y0.03103w Z Ž 8]12Hz.x q0.03901 w Z Ž 20]24Hz.x where ZŽ8]12 Hz. and ZŽ20]24 Hz. are ZŽ s, f . for the 8]12-Hz band and the 20]24-Hz band, respectively. A jackknifed validation procedure yielded strong correlations between the EEG measure and perfusion ŽFig. 6.. For the eyes-closed condition, a median correlation of 0.58 Žrange: 0.44]0.70. was found. The eyes-open condition yielded a lower correlation coefficient Žmedian s 0.45, range: 0.30]0.71.. 3.3. Detection of lateralized acti¨ ation We next compared the accuracy of PET, the cordance estimate of perfusion, and absolute and relative power measures in detecting lateralized activation over the motor strip. Both PET and cordance detected lateralized activation with accuracy exceeding chance levels of 0.10, while power did not exceed chance levels Ž P) 0.10. ŽTable 1.. 4. Discussion These results show that surface-recorded QEEG does reflect cerebral energy utilization in normal subjects, as evidenced by the moderately strong associations with relative perfusion. These 135 associations were stable across several conditions Ži.e. resting state vs. motor task, right-hand vs. left-hand activity. and across all brain regions, although the strength of the association was different in the eyes-open and eyes-closed conditions. In these normal subjects at most frequencies, there was a positive association between QEEG power and cerebral perfusion. The strength of the association varied significantly across the frequencies, with the strongest positive associations between QEEG measures and perfusion in these maximally alert subjects seen in a portion of the beta band between 20 and 28 Hz. This finding is consistent with the fact that beta activity above 20 Hz is characteristic of the activated state and is accentuated by visual attention ŽLopes da Silva et al., 1970., accurate task performance ŽFreeman and Van Dijk, 1988., and fine motor movements ŽMurthy and Fetz, 1992.. The only frequency bands in which there was a strong negative association between power and perfusion were those with a lower bound between 6 and 10 Hz, centering on the alpha band Ž8]12 Hz.. This finding is consistent with the fact that alpha activity is conceived of as ‘inhibitory’: it is characteristic of the resting state and is suppressed by arousal ŽSteriade et al., 1990. or activation during tasks ŽPfurtscheller and Neuper, 1992.. These findings also are consistent with the previous report which examined metabolism and alpha power in normal subjects, and found a negative association ŽBuchsbaum et al., 1984.. These results indicate that the association between electrical energy and cerebral perfusion differs between normal and diseased brain. The direction of the association between EEG power and cerebral perfusion in these normal subjects is opposite of that reported in previous studies of subjects with stroke or dementia, which reported a positive association between alpha power and perfusion, as well as a negative association between slow-wave power and perfusion ŽStigsby et al., 1981; Nagata et al., 1984, 1989; Nagata, 1988, 1989; Sloan et al., 1995; Valladeres-Neto et al., 1995.. These findings suggest that brain damage results in a fundamental alteration in the 136 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 coupling between slow-wave or alpha-band energy and perfusion in diseased brain. It is known that structural alterations in the brain may lead to a partial uncoupling of metabolism and perfusion ŽFink et al., 1996., even in brain tissue somewhat distant from the site of injury ŽMarchal et al., 1996.. Basic neurophysiologic studies highlight differences in the origins of normal and pathologic slow-wave activity. Normal slow waves are generated locally throughout the brain by layers II, III and V of the cerebral cortex, representing the firing of GABAergic interneurons inhibiting large pyramidal cells ŽBall et al., 1977; Steriade and Buzsaki, 1990. under the influence of thalamocortical projections ŽSteriade et al., 1990.. Pathologic slow waves are generated by these same layers, but in cells which have been altered by partial deafferentation Ži.e. a partial loss of modulating influences from the thalamus.. The partial loss of subcortical influences alters the nature of synaptic activity in layers II, III, and V of the cortex, and possibly the nature of the coupling between EEG power and perfusion. In addition to altering the nature of the associations between power and perfusion, brain damage also obscures the origin of slow waves in the EEG. Pathologic slow waves frequently do not arise locally, but instead represent a projected rhythm from distant Ždamaged. brain areas ŽSteriade et al., 1990.. Differences in the origins of normal and pathologic areas may alter or obscure the relationship between local brain electrical activity and metabolism or perfusion. This hypothesis is consistent with the findings of Valladeres-Neto et al. Ž1995., who examined the relationship between EEG activity and metabolism in patients with Alzheimer’s disease. They found that local EEG activity was much more strongly associated with hippocampal metabolism than with metabolism underlying the recording electrode. The negative association that we report between alpha frequency power and perfusion is the opposite of that reported in the presence of brain disease, but is entirely consistent with our understanding of the functional significance of alpha activity. Since this activity is inhibitory, it should be negatively associated with perfusion. It is pos- sible that the positive association seen between alpha activity and perfusion in diseased brain is peculiar to brain disease. Those areas that are capable of producing alpha activity are probably less affected by pathological processes and therefore have higher perfusion than areas in which alpha activity is not seen. The association between cordance and perfusion was stronger and more consistent than the association between power and perfusion. This reflects the fact that cordance combines measures of absolute and relative power, both of which have independent associations with perfusion ŽCook et al., 1998.. All scalp-recorded EEG waves represent the summed post-synaptic potentials of pyramidal cells firing in the cortex ŽSteriade et al., 1990., but we hypothesize that absolute and relative power are sensitive to different aspects of cortical cellular activity. Absolute power, for example, provides information about the energy produced by an ensemble of cells firing at a given frequency. Relative power additionally relates the energy from an ensemble of cells firing at a particular frequency to those firing at other frequencies, possibly providing information about the relative synchronization of cells under a given electrode or the relative number of cells firing at different frequencies. The cordance algorithm also entails an additional step of spatial normalization, relating the energy output from one electrode to all other electrodes on the scalp. There is insufficient knowledge about the basic neurophysiology of relative power to speculate further about the mechanisms which account for the enhanced association observed between cordance and perfusion. The current results indicate that we have identified two different ‘states’ of the cerebral cortex, based upon the association between absolute and relative power: concordance Žin which the mean absolute and relative power at that electrode are both above or below the respective global means. and discordance Žmean absolute power is below while relative power is above the respective global means, or vice versa.. These results show that electrodes in the concordant state, at least in the beta frequency band, are associated with higher perfusion than electrodes in the discordant state. A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 This finding is also consistent with our previous reports that show that in subjects with brain disease, discordance in the beta and theta frequency bands is seen over areas of brain dysfunction ŽLeuchter et al., 1994a,b.. It is important to note, however, that discordance is not synonymous with brain dysfunction, since all these normal subjects had a significant number of electrodes in the discordant state. Rather, concordance in most frequency bands is a categorical indicator seen at electrodes over brain areas in a higher relative perfusion state, and may be useful as an indicator of relatively enhanced brain activity. Conversely, discordance at most frequencies is a categorical indicator seen over areas in a lower relative perfusion state, possibly indicating relatively suppressed brain activity. Further studies to examine cordance in different cognitive, emotional, or treatment conditions could clarify this point. The current results also highlight the effects of technical factors in affecting the apparent relationship between EEG and perfusion measures. Some of the differences between the current results and those of previous studies may be explained by technical differences in recording and analysis methods. Many previous studies processed data from a referential montage ŽNagata et al., 1984, 1989; Nagata, 1988, 1989; Valladeres-Neto et al., 1995., which our previous work has shown produces substantially weaker associations with perfusion than the local Laplacian transformation ŽCook et al., 1998.. In other studies, associations were calculated between QEEG and perfusion data gathered from different brain regions ŽTolonen and Sulg, 1981; Wszolek et al., 1992., at different times ŽNagata et al., 1989; Jibiki et al., 1994., or while the subjects were in different mental states ŽNagata et al., 1984.. All of these factors have been demonstrated profoundly to influence brain activity Žcf. Posner et al., 1988.. Furthermore, the present study shows that the association between EEG and perfusion is consistently strongest in the beta bands, which frequently have not been examined by previous studies. It is important to note, however, that the usefulness of beta frequency measures was limited by the presence of artifacts, 137 which necessitated the elimination of several recordings from this study. The methods used in this study permitted us to detect associations between EEG power and perfusion that were significant across brain regions. It would be valuable to examine the topographic variability in these associations. A technical limitation of the current study is that the limited range of relative perfusion values Žwhich were clustered in a small range in these normal subjects. and the limited number of subjects, as well as deletion of electrodes because of artifacts, did not permit us to examine individual electrodes separately. Previous studies which examined subjects with brain disease found a broader range of perfusion values, which make correlations easier to examine. Although the associations in this study were significant for the brain as a whole, it is possible that some regions could show weaker or stronger associations than those reported overall. Future studies of normal subjects should examine possible regional variability in these QEEG measures. The methods of this report, aimed at the comparison of different modalities of brain imaging, are inherently fraught with difficulties and limitations. This problem is particularly marked when comparing methods as disparate as PET scanning Žbased upon tomographic reconstructions of brain. and QEEG Žmeasured from outside the scalp.. Even beyond the inherent technical limitations of this approach, no two imaging techniques ever can be entirely comparable. These techniques all contribute some similar and some unique information regarding brain function, and the correlational approach of this study only can estimate shared information. The goal of comparing two QEEG measures is less fraught with limitations. We previously reported that in patients with brain disease, cordance had stronger associations with cerebral perfusion Žmeasured with HMPAO-SPECT. than EEG power measures. The current study extends those findings and indicates that cordance has stronger associations with cerebral perfusion than QEEG power in normal subjects as well. The stronger associations seen with cordance may re- 138 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 flect the fact that cordance integrates absolute and relative power data, which are complementary measures of brain function ŽLeuchter et al., 1993.. The enhanced association with perfusion was most evident in the high-frequency beta band Žabove 20 Hz., a frequency range which is closely associated with mental effort and task activation. Furthermore, cordance detected the neurophysiologic processes of lateralized activation with greater accuracy than conventional EEG power measures. These findings suggest that cordance may offer particular advantages as a measure of cerebral activity in subjects without structural brain disease Acknowledgements This work was supported by research grant 1RO1 MH40705 and Research Scientist Development Award 1KO2 MH01165 from the National Institute on Mental Health ŽNIMH., and the Medication Development Research Unit contract 1YO1 DA50038 from the National Institute on Drug Abuse to the Department of Veterans Affairs ŽAFL., a NARSAD Young Investigator Award and training grant T32 MH17140 from the NIMH ŽIAC., and a fellowship from the Brookdale Foundation ŽROH.. The views in this manuscript represent those of the authors and do not necessarily represent those of the Department of Veteran Affairs. The authors gratefully acknowledge the expert consultation of Lynn Fairbanks, Ph.D., with the statistical analyses, as well as the assistance of Mariahn Smith, R.EEG.T., in the collection and processing of the EEG data, Michelle Abrams, R.N., in subject recruitment and scheduling, and Mychelle Blake, M.S.W., in the preparation of this manuscript. References Ball, C.J., Gloor, P., Schaul, N., 1977. The cortical electromicrophysiology of pathological delta waves in the electroencephalogram of the cat. Electroencephalography and Clinical Neurophysiology 43, 346]361. Berger, H., 1938. Das elektroenkephalogramm des menschen. Nova Acta Leop 6, 173]309. Buchsbaum, M.S., Kessler, R., King, A., Johnson, J., Cappelletti, J., 1984. Simultaneous cerebral glucography with positron emission tomography and topographic electroencephalography. In: Pfurtscheller, G., Jonkman, E.J., Lopes da Silva, F.H. ŽEds.., Brain Ischemia: Quantitative EEG and Imaging Techniques, Progress in Brain Research, vol. 62. Elsevier Science Publishers, New York, pp. 263]269. Cartheuser, C.F., 1988. Progressive hypoxia until brain electrical silence: a useful model for studying progressive interventions. Canadian Journal of Physiology and Pharmacalogy 66, 1398]1406. Cook, I.A., O’Hara, R., Uijtdehaage, S.H., Mandelkern, M., Leuchter, A.F., 1998. Assessing the accuracy of topographic EEG mapping for determining local brain function. Electroencephalography and Clinical Neurophysiology 107, 408]414. Des Rossiers, M.H., Kennedy, C., Patlak, C.S., 1974. Relationship between local cerebral blood flow and glucose utilization in the rat. Neurology 24, 389. Erecinska, M., Silver, I.A., 1989. ATP and brain function. Journal of Cerebral Blood Flow and Metabolism 9, 2]19. Fink, G.R., Pawlik, G., Stefan, H., Pietrzyk, U., Wienhard, K., Heiss, W.D., 1996. Temporal lobe epilepsy: evidence for interictal uncoupling of blood flow and glucose metabolism in temporomesial structures. Journal of Neurological Sciences 137, 28]34. Freeman, W.J., Van Dijk, B.W., 1988. Spatial patterns of visual cortical fast EEG during conditioned reflex in a rhesus monkey. Brain Research 422, 267]276. Gevins, A., 1990. Dynamic patterns in multiple lead data. In: Rohrbaugh, J., Johnson, R., Parasuraman, R. ŽEds.., Event-Related Brain Potentials: Basic Issues and Applications. Oxford University Press, New York, pp. 44]56. Hjorth, B., 1975. Technical contribution. An on-line transformation of EEG scalp potentials into orthogonal source derivations. EEG and Clinical Neurophysiology 39, 526]530. Iglesias, S., Marchal, G., Rioux, P., Beaudouin, V., Hauttement, A.J., de la Sayette, V., Le Doze, F., Derlon, J.M., Viader, F., Baron, J.C., 1996. Do changes in oxygen metabolism in the unaffected cerebral hemisphere underlie early neurological recovery after stroke? A positron emission tomography study. Stroke 27, 1192]1199. Ingvar, D.H., Sjolund, B., Ardo, A., 1976. Correlation between ¨ dominant EEG frequency, cerebral oxygen uptake and blood flow. EEG and Clinical Neurophysiology 41, 268]276. Ingvar, D.H., 1979. ‘Hyperfrontal’ distribution of the cerebral grey matter flow in resting wakefulness; on the functional anatomy of the conscious state. Acta Neurologica Scandinavica 60, 12]25. Ingvar, D.H., Risberg, J., 1967. Increase of regional cerebral blood flow during mental effort in normals and in patients with focal brain disorders. Experimental Brain Research 3, 195]211. Jibiki, I., Kurokawa, K., Fukushima, T., Kido, H., Yamaguchi, A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 N., Matsuda, H., Hisada, K., 1994. Correlations between quantitative EEG and regional cerebral blood flow ŽSPECT. in patients with partial epilepsy. Neuropsychobiology 30, 46]52. Klementavicius, R., Nemoto, E.M., Yonas, H., 1996. The Q10 ratio for basal cerebral metabolic rate for oxygen in rats. Journal of Neurosurgery 85, 482]487. Leuchter, A.F., Cook, I.A., Newton, T.F., Walter, D.O., Rosenberg-Thompson, S., Lachenbruch, P.A., Weiner, H., 1993. Regional differences in brain electrical activity in dementia: use of spectral power and spectral ratio measures. Electroencephalography and Clinical Neurophysiology 87, 385]393. Leuchter, A.F., Cook, I.A., Lufkin, R.B., Dunkin, J., Newton, T., Cummings, J.L., Mackey, K., Walter, D., 1994a. Cordance: a new method for assessment of cerebral perfusion and metabolism using quantitative electroencephalography. Neuroimage 1, 208]219. Leuchter, A.F., Cook, I.A., Mena, I., Dunkin, J., Cummings, J.L., Newton, T., Migneco, O., Lufkin, R., Walter, D., Lachenbruch, P., 1994b. Assessment of cerebral perfusion using quantitative EEG cordance. Psychiatry Research: Neuroimaging 55, 141]152. Lopes da Silva, F.H., Van Rotterdam, A., Storm van Leeuwen, W., Tielen, A.M., 1970. Dynamic characteristics of visual evoked potentials in the dog. II. Beta frequency selectivity in evoked potentials and background activity. EEG and Clinical Neurophysiology 29, 260]268. Marchal, G., Furlan, M., Beaudouin, V., Rioux, P., Hauttement, J.L., Serrati, C., de la Sayette, V., Le Doze, F., Viader, F., Derlon, J.M., Baron, J.C., 1996. Early spontaneous hyperperfusion after stroke. A marker of favourable tissue outcome? Brain 119, 409]419. Melamed, E., Lavy, S., Portnoy, Z., Sadan, S., Carmon, A., 1975. Correlation between cerebral blood flow and EEG frequency in the contralateral hemisphere in acute cerebral infarction. Journal of the Neurological Sciences 26, 21]27. Murthy, V.N., Fetz, E.E., 1992. Coherent 25]35 Hz oscillations in the sensorimotor cortex of the awake behaving monkey. Proceedings of the National Academy of Science USA 89, 5670]5674. Nagata, K., 1988. Topographic EEG in brain ischemia } correlation with blood flow and metabolism. Brain Topography 1, 97]106. Nagata, K., 1989. Topographic EEG mapping in cerebrovascular disease. Brain Topography 2, 119]128. Nagata, K., Mizukami, M., Araki, G., Kawase, T., Hirano, M., 1982. Topographic electroencephalographic study of cerebral infarction using computed mapping of the EEG. Journal of Cerebral Blood Flow and Metabolism 2, 79]88. Nagata, K., Tagawa, K., Hiroi, S., 1989. Electroencephalographic correlates of blood flow and oxygen metabolism provided by positron emission tomography in patients with cerebral infarction. Electroencephalography and Clinical Neurophysiolgy 72, 16]30. 139 Nagata, K., Yunoki, K., Araki, G., Mizumi, M., 1984. Topographic electroencephalographic study of transient ischemic attacks. Electroencephalography and Clinical Neurophysiology 58, 291]301. Obrist, W.D., Sokoloff, L., Lassen, N.A., et al., 1963. Relation of EEG to cerebral blood flow and metabolism in old age. Electroencephalography and Clinical Neurophysiology 15, 610]619. Okyere, J.G., Ktonas, P.Y., Meyer, J.S., 1986. Quantification of the alpha EEG modulation and its relation to cerebral blood flow. IEEE Transactions on Biomedical Engineering BME33]7, 690]696. Passero, S., Rocchi, R., Vatti, G., Burgalassi, L., Battistini, N., 1995. Quantitative EEG mapping, cerebral blood flow, and neuropsychological function in Alzheimer’s disease. Dementia 6, 148]156. Pfurtscheller, G., Neuper, C., 1992. Simultaneous EEG 10 Hz desynchronization and 40 Hz synchronization during finger movements. Neuroreport 3 Ž12., 1057]1060. Posner, M.I., Petersen, S.E., Fox, P.T., Raichle, M.E., 1988. Localization of cognitive operations in the human brain. Science 240, 1627]1631. Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T., 1986. Numerical Recipes: The Art of Scientific Computing. Cambridge University Press, Cambridge, England. Roy, C.W., Sherrington, C.S., 1890. On the regulation of the blood supply of the brain. Journal of Physiology ŽLondon. 11, 85]108. Serrati, C., Marchal, G., Rioux, P., 1994. Contralateral cerebellar hypometabolism: a predictor for stroke outcome? Journal of Neurology, Neurosurgery and Psychiatry 57, 174]179. Sloan, E.P., Genton, G.W., Kennedy, N.S., MacLennan, J.M., 1995. Electroencephalography and single photon emission computed tomography in dementia: a comparative study. Psychological Medicine 25, 631]638. Sokoloff, L., 1977. Relationship between physiological function and energy metabolism in the central nervous system. Journal of Neurochemistry 29, 13]26. Sokoloff, L., 1981. Relationships among local functional activity, energy metabolism, and blood flow in the central nervous system. Federation Proceedings 40, 2311]2316. Steriade, M., Buzsaki, G., 1990. Parallel activation of thalamic and cortical neurons by brainstem and basal forebrain cholinergic systems. In: Steriade, M., Biesold, D. ŽEds.., Brain Cholinergic Systems. Oxford, New York, pp. 3]62. Steriade, M., Gloor, P., Llinas, R.R., Lopes da Silva, F.H., Mesulam, M.M., 1990. Basic mechanisms of cerebral rhythmic activities. Electroencephalography and Clinical Neurophysiology 76, 481]508. Stigsby, B., Johannesson, G., Ingvar, P.H., 1981. Regional EEG analysis and regional cerebral blood flow in Alzheimer’s and Pick’s diseases. Electroencephalography and Clinical Neurophysiology 51, 537]547. Toga, A.W., Cannestra, A.F., Black, K.L., 1995. The tem- 140 A.F. Leuchter et al. r Psychiatry Research: Neuroimaging Section 90 (1999) 125]140 poralrspatial evolution of optical signals in human cortex. Cerebral Cortex 6, 561]565. Tolonen, U., Sulg, I.A., 1981. Comparison of quantitative EEG parameters from four different analysis techniques in evaluation of relationships between EEG and CBF in brain infarction. Electroencephalography and Clinical Neurophysiology 51, 177]185. Valladeres-Neto, D.C., Buchsbaum, M.S., Evans, W.J., Nguyen, D., Nguyen, P., Siegel, B., Stanley, J., Starr, A., Guich, S., Rice, D., 1995. EEG delta positron emission tomography and memory deficit in Alzheimer’s disease. Neuropsychobiology 31, 173]181. Wszolek, Z.K., Herkes, G.K., Lagerlund, T.D., Kokmen, E., 1992. Comparison of EEG background frequency analysis, psychologic test scores, short test of mental status, and quantitative SPECT in dementia. Journal of Geriatric Psychiatry and Neurology 5, 22]30.