dynamic spectral analysis findings in first episode and chronic schizophrenia

223

Intern. J. Neuroscience, 116:223–246, 2006Copyright 2006 Taylor & Francis Group, LLCISSN: 0020-7454 / 1543-5245 onlineDOI: 10.1080/00207450500402977

DYNAMIC SPECTRAL ANALYSIS FINDINGS INFIRST EPISODE AND CHRONIC SCHIZOPHRENIA

ANTHONY HARRIS

Brain Dynamics CentreWestmead HospitalWestmead, Australia

and

Discipline of Psychological MedicineUniversity of SydneySydney, Australia

and

Department of PsychiatryWestmead HospitalWestmead, Australia

DMITRIY MELKONIAN


and

Department of PsychiatryWestmead HospitalWestmead, Australia

Received 4 May 2004.Address correspondence to Dr. A.W.F. Harris, Brain Dynamics Centre, Westmead Hospital, West-

mead, NSW 2145, Australia. E-mail: [email protected]

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.

224 A. HARRIS ET AL.

LEANNE WILLIAMS


and


and

School of PsychologyUniversity of SydneySydney, Australia

EVIAN GORDON


and


and

Brain Resource CompanyUltimo, Australia.

The quantified analysis of the electroencephalogram (qEEG) has enabled theextraction of additional psychophysiological information from the raw EEG,but in turn has introduced a number of distortions. This study compared Dy-namic Spectral Analysis (DSA), a novel and mathematically stringent tech-nique for the evaluation of qEEG activity with conventional power spectralanalysis in subjects with both first episode and chronic schizophrenia and matchedcontrols. Advantages of the technique in the automated processing of data,rejection of artefact, avoidance of artefact introduced by the mathematical trans-formation of the data and the identification of irregular low frequency artefactualactivity “pi” are discussed in detail. Using this method, the study has confirmed

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.

DSA FINDINGS IN SCHIZOPHRENIA 225

past observations of increased slow wave activity in schizophrenia, and identi-fied a decrease in peak frequency in the alpha band in the subjects with chronicschizophrenia. The two clinical groups differed in mean peak frequency in thedelta band with the first episode schizophrenia subjects having a raised meanpeak frequency and the subjects with chronic schizophrenia having a loweredmean peak frequency. The results suggest continued change in the EEG withillness chronicity in schizophrenia. These changes were most evident in thefrequency domain emphasizing the importance of routine measurement of meanband frequencies in qEEG studies.

Keywords alpha, delta, dynamic spectral analysis, EEG, first episode, schizo-phrenia

INTRODUCTION

The use of quantified electroencephalography (EEG) of raw EEG to assessand measure central nervous system activity enables the extraction of addi-tional psychophysiological information from the raw EEG (Gevins, 1998).However, the added spectral information gained needs to be balanced by afull appreciation of methodological adequacy of the techniques. A majorlimitation of the power spectrum estimation is that this technique is addressedto stationary random processes (Blackman & Tukey, 1959) but the activitypatterns of EEG are typically both highly irregular and non-stationary. Thatis, their statistical character changes slowly or intermittently as the result ofongoing activity of the sources of scalp potentials (Barlow 1985).

A major approach to overcoming this inconsistency relies on the conceptof local stationarity (Barlow, 1985). This assumes that although the EEGrecording is inherently nonstationary, within small intervals of time the pro-cess departs only slightly from stationarity. This necessitates segmenting theEEG into short sections within which the data may be considered as quasi-stationary. A mandatory aspect of this approach is the removal of EEG seg-ments contaminated by artefacts caused by myogenic or ocular movements(Pivik et al., 1993). The choice of artefact-free epochs is based on the screen-ing of records or the automated artefact rejection procedures (Creutzfeldt etal., 1985). Because the segmentation is adaptive, the EEG segments selectedfor analysis are relatively short and may have different length. This createsdifficult computational problem in numerical Fourier analysis.

Historically, widespread applications of the frequency domain methodsin EEG scientific and clinical research has been facilitated by the fast Fouriertransform (FFT) algorithm that provides an effective numerical estimation of

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


digital Fourier transform. Unfortunately, the FFT is not suited for local spec-tral decomposition of the signal in short-time intervals. This limitation hasled numerous authors to design special procedures of short-time EEG spectralanalysis often labeled by the term “dynamic spectral analysis” (DSA) (Florian& Pfurtscheller, 1995; Melkonian et al., 1998).

Dynamic spectral analysis (DSA) uses the similar basis functions (SBF)algorithm (Melkonian, 1987; Melkonian et al., 1998) and relies on classicalmethods of numerical estimation of trigonometric integrals based on the polynomialexpansion of the function to be transformed (Filon, 1928). The algorithmdecomposes a piecewise linear approximation of the signal into the sum oftypical finite elements with a simple analytical form of a frequency spectrum.Dynamic spectral analysis of brain potentials developed on this basis (Melkonianet al., 1998) provides a means for the short-time spectral decompositions ofEEG signals and significantly improve the accuracy of frequency domainmeasures compared to FFT, an advantage of the polynomial technique notedin previous research (Schütte, 1981; Mäkinen, 1982)

It is important to note that DSA using SBF changes the theoretical frame-work from discrete Fourier transform to classical Fourier integral transform,which deals with continuous functions of the frequency and time. Two of thereasons why the DSA of biological signals supported by Fourier integrals isconsidered to be superior to FFT based techniques are (Harris, 1998): (i)DSA provides continuous Fourier spectrum instead of discrete power spec-trum relevant to the FFT and (ii) DSA preserves the phase components thrownaway in the power spectrum estimation. This ability of DSA to provide thefrequency domain equivalent of the time domain signal leads to deeper under-standing of EEG frequency domain parameters, making this already usefulmeasure more informative. With respect to artefact introduced by the fre-quency transformation of data with an initial or final nonzero value (Nuwer,1988), the algorithm allows for unequal sampling and adaptive segmentationof the EEG. This removes the need for the use of standard windows in EEGspectral analysis.

The aim of this study was to examine the EEG, using DSA as well asconventional FFT, in both first episode and chronic schizophrenia, as againstmatched normal controls to see if the new technique was comparable toresults using conventional FFT and to assess the benefits arising from the useof the new technique. Secondly, the study examines differences between thegroups in peak frequency in the different bands, as decreased EEG bandfrequency has been associated with changes in cortical activity and function(Klimesch et al., 1993; Klimesch et al., 1996). Changes in peak frequency

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


may be a more accurate means of assessing the decrement in cortical func-tioning found in schizophrenia. Subjects with schizophrenia were hypoth-esized to have greater slow wave activity and slower mean peak frequenciesin delta and theta, and a slower mean peak frequency in alpha.

METHOD

Subjects

Patients with Chronic Schizophrenia. Forty subjects with chronic schizo-phrenia aged between 23 and 51 years of age (male n = 26; female n = 14),with a mean age of 36.0 years (SD = 7.1 years), were recruited from hospitaland community health centers (see Table 1). Subjects were interviewed usinga semi-structured interview schedule and were questioned about their previ-ous psychiatric history, family psychiatric history, medical history, and levelof educational attainment. Subjects were excluded on the basis of a recenthistory of substance abuse, or past history of substance dependence, epilepsy,other neurological disorders, mental retardation, or head injury. Subjects werepredominant right handed.

Diagnosis was confirmed using Section G (schizophrenia and psychoticdisorders) of the Composite International Diagnostic Interview (World HealthOrganization, 1992) or by consensus diagnosis of fully qualified psychiatrists

Table 1. Clinical variables of subjects with schizophrenia

First episode Chronic schizophrenia schizophrenia

(n = 40) (n = 40)

Mean SD Mean SD p values

Age (yrs) 19.6 3.3 36.0 7.1 0.000Illness duration (yrs) 1.1 1.3 14.3 7.0 0.000Chlorpromazine equi 253 197 520 423 0.001

PANSS scoresPositive symptoms 17.9 7.3 20.3 6.5 0.045Negative symptoms 19.3 6.1 20.6 6.5 n.s.General symptoms 38.6 8.2 37.6 8.3 n.s.PANSS Total 75.8 15.8 78.5 18.7 n.s.

n.s. = nonsignificant.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


according to the DSM-III-R (Diagnostic and Statistical Manual of MentalDisorders, 3rd edition revised; American Psychiatric Association, 1987). Allsubjects with chronic schizophrenia had been unwell for a minimum periodof at least four years (mean duration of illness 14.3 yrs, SD = 7.0 yrs). Allwere medicated. Twenty subjects were on typical antipsychotics, mostly de-pot preparations, 7 were on atypical antipsychotics and 13 were on clozapine.The average dose of medication (Lambert, 1998, 1999) was 520 chlorprom-azine equivalents (SD = 423 cpz equi).

Patients with First Episode Schizophrenia. Forty subjects (male =28, female = 12) with their first episode of schizophrenia aged between theages of 13 and 25 years (mean = 19.6 yrs; SD = 3.2 yrs) were tested. Firstepisode schizophrenia subjects were defined as those young people from theproject with psychotic symptoms that warranted a diagnosis of either schizo-phrenia or schizophreniform disorder. Diagnosis was made by means of aconsensus conference of fully qualified psychiatrists according to DSM-IV(Diagnostic and Statistical Manual of Mental Disorders, 4th edition; Ameri-can Psychiatric Association, 1994) criteria.

The majority of subjects were medicated with atypical antipsychoticsalone (mean dose 250 chlorpromazine equivalents; SD = 202 cpz equi), al-though a small number were also medicated with antidepressant or anticho-linergic medications. Four subjects were on no medication. The average du-ration of symptoms prior to presentation was 12.5 months (SD = 16.4 months).

The Positive and Negative Syndrome Scale (PANSS—Kay et al., 1986)was the principal scale used to rate individual symptoms for both patientgroups. This was administered to the subjects by psychiatrists or psychologistsof at least Master’s level trained in the use of the instrument and who hadreached acceptable levels of inter-rater reliability (correlation coefficient > 0.8)or had reached similar levels of inter-rater reliability in trials with the author.

Comparison of First Episode and Chronic Schizophrenia Subjects. Acomparison of subjects with first episode and chronic schizophrenia (see Table1) was performed using the Mann-Whitney U statistic. The group with chronicschizophrenia had more positive symptoms (U = 592.500, N

1 = 40, N

2 = 40, p =

.045) and were treated with higher doses of medication (U = 414.500, N1 = 40,

N2 = 38, p = .001) but did not differ on measures of negative symptoms.

Normal Control Subjects. Control subjects for the group of subjects withchronic schizophrenia were recruited from the surrounding community or

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


from non-clinical hospital staff. All normal controls were questioned abouttheir previous medical history and excluded on the basis of a recent historyof substance abuse, or past history of substance dependence, epilepsy, otherneurological disorders, mental retardation, psychiatric disorder including pastAttention Deficit Hyperactivity Disorder, disruptive behaviour disorders, orlearning disabilities. Handedness was checked using questions from the Edin-burgh Handedness Inventory.

Control subjects were age and sex matched to within 2 years to subjectsunder the age of 25 years and to within 5 years for subjects over the age of25 years. The closer age matching of the younger subjects reflected the matura-tional changes in the EEG that are to be expected up to early adulthood(Niedermeyer, 1999). The average age of the young normal control groupwas 19.7 yrs (SD = 3.9 yrs) as against an average age of 36.6 yrs (SD = 7.3yrs) for the old normal control group. This did not differ from the age matchedgroups with schizophrenia.

Voluntary consent was obtained from all subjects according to NationalHealth and Medical Research Council guidelines.

Data Acquisition

EEGs were acquired as part of an electrophysiological battery. The EEGswere recorded using an electrocap (Blom & Anneveldt, 1982) from 19 siteswith the sites Fz, Cz,and Pz being reported here from the eyes closed condi-tion. Linked earlobes served as a reference. Both horizontal and vertical EOGactivity was recorded with two bipolar electrodes placed 1 cm lateral to theouter canthus of each eye and two bipolar electrodes placed above and belowthe center of the right eye. All electrode impedances were maintained at 5kOhms or less throughout the recording. During the procedure subjects wereseated in a comfortable chair and asked to limit any eye movement.

All potentials were acquired on a Syn Amps (NEURO SCAN Inc) 32channel DC system with a gain of 10,000, a digitalization rate of 250 Hz,and an upper bandpass filter set at 50 Hz. A continuous EEG recording wasacquired until 2 min of EEG was collected. All data was stored digitally forfurther analysis.

Artefact Rejection

Selection of quasi-stationary segments of EEG signals for spectral estimationwas based on automated identification of artefact contaminated EEG sections.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


A difficult aspect of this problem is that EEG contamination usually comesfrom different sources, relating to (i) extraneous electrical activity from theperson eg muscle artefact and (ii) artefact secondary to data processing. Thisnecessitates a number of methods to identify and eliminate artefact contami-nated segments.

A widely accepted approach to artefact identification associates artefactwith EEG data where the absolute maximum amplitude of the signal exceedsa fixed threshold (Agarwal et al., 1998). The estimation of absolute maxi-mum amplitude is illustrated in Figure 1, where the bold solid line 1 is a rawEEG signal and M is the absolute maximum amplitude. An original aspect ofthis approach to the amplitude threshold estimation is that the authors sepa-rate the raw EEG signal into 2 major components:

s(t) = π(t) + x(t), (1)

where s(t) is a raw EEG signal, x(t) is the EEG signal expanding over con-ventional frequency bands of delta, theta, alpha and beta, π(t) is a nonstationarylow frequency component named “pi.” The pi is extracted from the signalusing an equal weight digital boxcar filter (Cook & Miller, 1992) with thelength of window 180 ms. This technique carries out adaptive filtering, that

Figure 1. Estimation of EEG absolute maximum amplitude using the baseline. Bold solidcurve 1 shows raw EEG signal whose absolute maximum amplitude is M. The solid line isa baseline [irrelevant low frequency component pi : π(t)]. Bold solid curve 2 shows EEGsignal measured relatively to the baseline (removal of π(t) from raw signal). Absolute maximumE now selectively reflects maximum amplitude for conventional EEG bands.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


is, the frequency content of both EEG and pi varies depending on the mo-mentary properties of raw EEG signal, muscular, skin conductance, electrical,and so on artefacts. Note that physically pi serves as baseline for extractionof conventional EEG. In any fixed EEG segment, each of these componentsis estimated separately. The EEG signal is estimated by the absolute maxi-mum amplitude, M. The π is estimated by the parameter, B = π

max – π

mean/

T, where πmax

and πmean

are the maximum and mean of π(t) in the interval ofinterest and T is the length of the interval. Following Agarwal and colleges(1998), the maximum amplitude threshold for raw EEG was defined as 300µV. The threshold values of new measures, regarding as indicative of artefactcontaminated EEG segments were as follows: M = 45 µV and B = 100 µV/s.

Another method of detecting artefact-contaminated EEG is to eliminateall segments with abnormal patterns of EEG activity. This approach utilizeshigh resolution fragmentary decomposition (FD) (Melkonian et al., 2003).This model-based method of non-stationary electrophysiological signal analysisprovides a means for the estimation of the fine temporal structure of EEGwaveforms. FD defines zero-crossings of EEG signal [x(t) in terms of equa-tion (1)]. On this basis, the automated search of an artefact-free interval [a

n,

bn] is additionally specified by the following two conditions: (i) ZC

MAX < 0.25

s and (ii) Nz

< 50, where ZCMAX

is the maximum value of the intervals be-tween subsequent zero-crossings in [a

n, b

n] and N

z is the number of zero-

crossing. These conditions remove from the analysis segments which containtoo slow or too fast activities that contaminate characteristic patterns of EEGactivities.

Periodogram Method of EEG Spectral Analysis

EEG spectral estimates were obtained using the conventional periodogrammethod. Algorithms for this method are based on a classical definition of thepower spectrum, P( f ), of the stationary random process, X(t) (Blackman &Tukey, 1958, p. 7):

P( f ) = lim T x(t) ⋅ e–12πft dt2

, (2)

where t and f are the time and frequency, respectively, i = √ – 1 and Tdefines the limits of integration.

Although an exact concept of the power spectrum presumes infinitelimits of integration, any technique of spectral measurements relies on a limited

1

T→∞

}

T/2

–T/2

�

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


number of x(t) segments of finite length. This reduces major computationalproblems to numerical computation of finite Fourier integrals of the form

X( f ) = L

x(t) ⋅ e–12π ft ⋅ dt = XC( f ) – iX

S( f ), (3)

where a and b define particular integration interval, L = b – a and XC( f ) and

XS( f ) are real and imaginary parts of the complex spectrum X( f ), that is,

finite cosine and sine Fourier transforms defined by

XC( f ) =

Lx(t) cos 2πftdt, X

S( f ) =

Lx(t) sin 2πftdt. (4)

The amplitude spectrum, M( f ), is

M( f ) = X( f ) = √X2C

( f ) + X2S( f ), (5)

Given that EEG is contaminated by different artefacts, it is essential thatEEG be broken into artefact free sections. According to most clinical studiesthe epoch of EEG analysis for reliable spectrum estimation must be relativelylong, for example, 60 s (Pivik et al., 1993). Hence, if a typical length of anartefact free segment is from 1 to 3 s, the spectral measures should be col-lected from tens of EEG segments. This provides segment amplitude spectra,M

1( f ), . . . , M

N( f ), where N is the number of segments. The EEG amplitude

spectrum is simply the average of these characteristics defined by

ME( f ) = N M

i( f ). (6)

Since the equation (2) defines the power spectrum as the square of am-plitude spectrum, the EEG power spectrum is

PE( f ) = M 2

E( f ). (7)

Two spectrum analysis techniques were employed for estimation of Fou-rier transforms: conventional FFT and DSA using the SBF algorithm.

Conventional FFT

There are two major theoretical approaches that support the application ofFFT to the EEG power spectrum estimation. The first defines EEG spectral

1 �b

a

b

a��b

a

1 1

1�N

i =1

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


power in terms of discrete Fourier transform, that is, the EEG is primarilyregarded as a time series, not as a continuous process. Under this frameworkthe FFT is an adequate algorithm for the estimation of discrete Fourier trans-form. The second approach relies on a classical definition of power spectrumin terms of continuous time and frequency functions (equation 2), and re-gards FFT as computational tool that provides approximate estimation offinite Fourier integrals having the form (2) or (3).

The computational outputs,, as well as limitations posed by the FFTtechniques of both approaches, are the same. The FFT is strictly applicableonly to time series with 2n evenly spaced samples. Given 32, 64, and 128point time series, the durations of corresponding EEG intervals (samplinginterval = 4 ms) are 1.28 s, 2.56 s, and 5.12 s, respectively. This studyemployed a 64 point time series. EEG records were screened and dividedinto artefact free segments of 2.56 s duration and subjected to a Fast FourierTransform. Power spectra were estimated by averaging over 40 segments.The spectral power values in frequency increments of 0.391 Hz were summedin the following frequency bands: delta (1.0–3.0 Hz), theta (4.0–7.0 Hz),alpha (8.0–13.0 Hz), alpha 1 (8.0–9.0 Hz), alpha 2 (10.0–13.0 Hz), and beta(14.0–30.0 Hz). Natural log transformations were undertaken for the result-ing band-related absolute power values, for each frequency band and midlinesite under “eyes closed” conditions, to control for skewness.

DSA using Similar Basis Functions (SBF) Algorithm

Similar basis functions (SBF) algorithm is the computational technique thatestimates finite Fourier integrals (2) using piecewise linear approximation oftransformed function x(t), that is, a first order polynomial expansion. Giventhe signal samples, x

n = x(t

n), at N time points t

n (n takes values from 0 to

N – 1), the criterion to find the approximation function, h(t), is the interpola-tion condition:

hn = x

n for n = 0, 1, . . . , N – 1, (7)

where hn = h(t

n).

Figure 2A exemplifies the principle of piecewise-linear approximation.Given x(t) (bold line), the approximating function, h(t), is the broken linecreated by joining consecutive samples 0, 1, 2, 3, and 4 (time points t

0, t

1, t

2,

t3, and t

4, respectively) by straight lines (in the same fashion the approxima-

tion can be continued for any number of following nodal points, i.e., t5, t

6

, . . .).

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


Effective computational scheme of the SBF algorithm, adopted form thefinite-element methods, is based on the conversion of the approximating functioninto the weighted sum

h(t) = anr

n(t) = a

nr (t/t

n+1), (8)

A

2

B

h

g

o

f

i

a

e

b

d

c

1

0

3

4

0

0.2

0.4

0.6

0.01 1 100 frequency

1

2

D

0

0.5

1

0 1

C

Figure 2

Figure 2. (A) Bold solid curve exemplifies transformed time function. The piecewise-lin-ear approximating function (solid line with the breaks at nodal points 0, 1, 2, 3, and 4) isconverted into the sum of triangles (see equation 8) with the peaks at the points 0, 1, 2, and3. (B) Shows that the triangle abc is the sum of TBEs ohc,oga and odf. (C). Shows TBE.(D) Real (curve 1) and imaginary (curve 2) parts of TBE’s complex spectrum.

N–1

n = 0� �

N–1

n = 0

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


where an is weighting coefficient, r

n(t) is a similar basis function and r(t) is

triangular basis element (TBE) shown in Figure 1C. The similarity relation-ship, r

n(t) = r(t/t

n+1), following from (8), shows that each SBF is a scaled

copy of TBE.The geometric principle explaining equation (8) may be seen by refer-

ence to Figure 2A and B. Figure 2A links a triangle with each of samplesdenoted 0, 1, 2, and 3. The first triangle (sample 0) belongs to the categoryof SBF and corresponds to the first term in (8) under n = 0. A general formof remaining triangles is reflected by the triangle abc, shown in Figure 1B.The figure shows that abc is the sum of triangles oga, ofd, and ohc each ofwhich belongs to the category of SBF. The number of SBF with differentbases is equal to the number of break points of approximating function.Accordingly a relatively simple formulae estimate the weighting coefficientsa

n in (8) from h

n samples (Melkonian et al., 1998). Now that the approximat-

ing function is converted into the weighted sum of similar basis functions,the article turns now to the frequency characteristics of TBE,

RC

( f ) = r(t) ⋅ cos 2π ft ⋅ dt = 1 – cos 2π f ,

and

RS( f ) = r(t) ⋅ sin 2π ft ⋅ dt =

2πf – sin 2π f ,

Based on the scaling theorem applied to the similarity relations, thecosine and sine Fourier transforms of r(t/t

n), are t

nR

C(t

nf ) and t

nR

S(t

nf ), re-

spectively. Therefore, computations of the spectral characteristics of h(t) arereduced to standard manipulations with relatively simple frequency charac-teristics of the TBE.

As a means of testing computational accuracy and frequency resolution,the authors employed the SBF algorithm to recalculate the initial EEG seg-ments via inverse Fourier transform of spectral characteristics (imaginary orreal part of the complex spectrum) to the time domain. Computer experi-ments using this combined technique of direct and inverse Fourier transformshave shown that estimation of 100 discrete points per decade (logarithmicfrequency scale) provides a correct representation of frequency profiles ofspectral characteristics in the following conventional frequency bands: delta(1–4 Hz), theta (4–8 Hz), alpha (8–13 Hz) and beta (13–30 Hz). The spectralcharacteristics of the EEG segments have been computed for the frequencyrange from 1 to 30.2 Hz at frequencies f

j = f

1cj from j = 0 to j = 148, where

1

0

1

0

�

�

(2π f)2

(2π f)2

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


f1 = 1 Hz and c = 1.02329 (this parameter provides 100 samples per decade).

This provided 149 equidistant points in the logarithmic frequency scale dis-tributed between the bands as follows: delta—61 points, theta—30 points,alpha—21 points, and beta—37 points.

Amplitude and power spectra of the EEG record were estimated for eachsubject from equations (6) and (7) using 60 EEG segments. Given a separateEEG segment, the problem arises of whether a peak in the band of interestreflects the presence of a true rhythmic component or is simply caused by therandom fluctuations of the underlying process. In this respect the advantageof equations of (6) and (7) is that averaging of spectral characteristics fromstandard number of EEG segments effectively emphasizes repeated spectralpatterns, that is, the peaks produced by rhythmic EEG components.

The peak amplitude and peak frequency of the dominant spectral com-ponent in a defined frequency band has been estimated as the parameters(peak amplitude and peak frequency) of the maximum peak identified fromthe amplitude spectrum within a corresponding frequency range. The ampli-tude spectrum was chosen for the peak analysis because the peak amplitudederived from this characteristic may be directly associated with the amplitudeof the corresponding frequency components. By contrast, the squared magni-tude (power) spectrum is a nonlinear transformation that violates relation-ships between distinct frequency components introduced by the Fourier analysis.The area of the amplitude spectrum within each of frequency bands wasestimated. This parameter is useful in cases where spectral characteristics donot show clear pronounced peaks for some frequency bands.

Statistics

Before analysis, all EEG component values were natural log transformed soas to satisfy the normal distribution requirements of parametric statisticaltests. Outliers, as defined as 2.5 standard deviations from the mean values forthe group, were removed and their values replaced by group means.

Using SPSS (1999) for examination of within-subject effects, each powerwas submitted to a two way MANOVA model over the factors of group (firstepisode schizophrenia/chronic schizophrenia, young/old normal controls) andsite (Fz, Cz, Pz). Greenhouse-Geisser correction was used to adjust the re-sults for violation of Mauchly’s test of sphericity; however, only sphericityassumed degrees of freedom are reported. Average between-subject effectswere tested using a two-way ANOVA within the General Linear Model (SPSS,1999). Results were first tested for violation of Levene’s test of error variance.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


Post hoc comparisons were performed using two-way t-tests for the group bysite comparisons to better define the region of significant difference. Forresults that had not been hypothesized a correction for multiple comparisonsto a level of 0.015 was applied (Keppel, 1982).

RESULTS

Some Properties of Irrelevant Low Frequency Activity

Properties of low band irrelevant activity (pi) were tested using characteristicrecordings of raw EEGs from different subjects and recording sites. Figure3A illustrates a characteristic 80 s segment of pi activity that does not exceeda fixed 300 µV threshold and consequently can be accepted by conventionalartefact rejection criterion (fixed maximum amplitude threshold) as an “artefact

Figure 3. (A) Low band irrelevant activity (pi) in raw EEG. Note marked variability. (B)Amplitude Spectrum demonstrating pi activity in delta band.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


free” section of EEG activity. However, the estimate of the amplitude spec-trum of this process, illustrated by Figure 3B, shows the presence of thespectral components of pi activity in the delta band. Although the pi compo-nents in the delta band are significantly smaller than amplitudes of multiplefrequency peaks in lower frequencies, they are comparable with the frequencycomponents of the delta EEG signals (note that deflections of pi are signifi-cantly larger than normal deflections of EEG).

A quantitative characterization of pi activity is difficult due to its highlyirregular and non-stationary character that affects both the time and frequencydomain parameters of this process. We were unable to find any regularity inthe distributions of the peak frequencies of pi components.

Subjects with Chronic Schizophrenia vs. Old Control Subjects

Dynamic Spectral Analysis. A significant difference was observed betweenchronic schizophrenia and their age and gender matched controls, for peakfrequency (see Table 2) in the delta (F = 6.862; df = 1,78; p = .011) andalpha bands (F = 7.343; df = 1,78; p = .008), with the subjects with chronicschizophrenia having a higher mean frequency in the delta band and a lowermean frequency in the alpha band. In the examination of group by site inter-action a significant difference was observed for the delta band alone (F =10.298; df = 2,156; p = .000). Post hoc analysis of this interaction found asignificant difference at both Fz (t = 3.209; df = 78; p = .002) and Cz (t =3.367; df = 78; p = .001) with the subjects with chronic schizophrenia havinga higher mean frequency at both sites. This was reversed at Pz with the

Table 2. DSA peak frequency between group significant differences at midline sites

Peak frequency

First episode schizophrenia Chronic schizophrenia vs. young controls vs. old controls

Between Gp Gp × site Between Gp Gp × site

Delta 0.008 — 0.011 0.001Theta — — — —Alpha — — 0.008 —Beta — — — —

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


control subjects having a higher mean delta frequency, however the differ-ence was not significant.

When the two groups were compared on measures of amplitude spec-trum (see Table 3) a different though complementary picture emerged. Thegroup with chronic schizophrenia had significantly more delta (F = 7.722;df = 1,78; p = .007) and theta (F = 17.290; df = 1,78; p = .000) band activity.When the group by site interaction was examined no significant differencewas observed.

Conventional Power Spectral Analysis. When subjects with chronic schizo-phrenia were compared to their age and sex matched control subjects, signifi-cant differences were observed in the delta (F = 18.050; df = 1,78; p = .000)and theta (F = 21.276; df = 1,78; p = .000) bands. When the interaction ofgroup by site was examined both theta (F = 3.623; df = 2,156; p = .039) andalpha 2 (F = 4.162; df = 2,156; p = .036) power bands were significant. Post-hoc analysis of this result found that the subjects with chronic schizophreniahad significantly higher theta power values at Fz (t = 4.113; df = 78; p =.000), Cz (t = 4.459; df = 78; p = .000) and Pz (t = 4.733; df = 78; p = .000)and the age matched controls had higher alpha 2 values at Pz alone (t = –2.109; df = 78; p = .038).

Table 3. Comparison results for conventional Power Analysis (cFFT) and DynamicSpectral Analysis (DSA) Amplitude Spectrum between group significant differences atmidline sites

Power/amplitude

First episode schizophrenia Chronic schizophrenia vs. controls vs. controls

cFFT DSA cFFT DSA

Between Gp × Between Gp × Between Gp × Between Gp ×Gp site Gp site Gp site Gp site

Delta 0.050 0.000 0.007Theta 0.000 0.039 0.000Alpha 1 —Alpha 2 0.036Alpha TBeta

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


Subjects with First Episode Schizophreniavs. Young Control Subjects

Dynamic Spectral Analysis. In the group with first episode schizophreniathe significant results were restricted to the delta band alone. The subjectswith their first episode of schizophrenia had a lower mean frequency (F =7.360; df = 1,76; p = .008) and a greater amplitude spectrum than theircontrol subjects (F = 3.959; df = 1,76; p = .050).

Conventional Power Spectral Analysis. No significant results were observed.

DISCUSSION

The main findings reported in this article depend on the methodological in-novations of dynamic spectral analysis and the concepts of fragmentary

Figure 4. Group average amplitude spectrum (Pz; eyes closed). The relative increase in slowwave activity (1–7 Hz) is evident for both the first episode and the chronic schizophreniagroups of subjects as against their normal control groups. The slowing of alpha peak fre-quency in the chronic schizophrenia subjects can be readily seen. (See Color Plate I at endof issue.)

0

0.4

0.8

1.2

1.6

1 10 100f, Hz

µV

First episode Chronic Young NC

Old NC

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


decomposition of non-stationary EEG. Most quantitative approaches to EEGfrequency analysis are based on simplifications that do not account for EEGnon-stationarities. However, EEG frequency bands not only significantly varybetween different individuals but also depend on the momentary propertiesof EEG signals, particularly on the possible relationships between neuralsystems producing different frequency components. Using technique of digi-tal adaptive filtering, the authors found that raw EEG signal contains irrel-evant low frequency activity, pi, which does not possess information associ-ated with classical EEG bands. An important feature of pi is that this processhas an extremely irregular character and may remain at a practically zerolevel for tens and hundreds milliseconds. Such kind of behavior is quitedifferent from conventional patterns of EEG activity. The origins of pi areunclear. This activity may reflect changes in skin conductance. Another pos-sible source is glial activity given that the frequency content of glial depolar-ization belongs to significantly lower frequency bands than relevant electricalactivity of cortical neurons (Roitbak et al., 1987). This same irregular lowfrequency activity is likely to distort EEG quantitative analysis in the deltaband. Its removal allows for a more accurate estimation of delta activity.

This study demonstrated the compatibility of DSA amplitude spectrumresults with those found using conventional power spectral analysis and in-deed suggests their greater sensitivity. The overarching finding is that in-creased delta band activity remains the essential difference between age matchedgroups of subjects with schizophrenia and the normal population, but thatthis is most pronounced for subjects with chronic schizophrenia. In this theresults support those of numerous other studies (Stevens & Livermore, 1982;Mukundan, 1986; Guenther et al., 1988; Karson et al., 1988; John et al.,1994; Sponheim et al., 1994; Harris et al., 1997).

In the group with chronic schizophrenia, the increase in delta amplitudespectrum was accompanied by an increase in theta amplitude spectrum. Al-though there is a propensity for antipsychotic medication to raise theta power(Itil, 1977; Westphal et al., 1990; Knott et al., 2001), increased theta powerhas also been observed in unmedicated subjects with schizophrenia (Stevens& Livermore, 1982; Karson et al., 1987; Westphal et al., 1990; Harris et al.,1997; Knott et al., 2001). This increase in slow wave activity saw an increasein delta mean frequency which suggests that a downward shift of the EEGspectrum may have occurred in chronic schizophrenia. This has also beenobserved in another population of subjects with chronic schizophrenia whohad recently been started on clozapine (Knott et al., 2001). A third of thesubjects in the group with chronic schizophrenia were medicated with clozapine.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


Knott and colleagues (1997) suggested that this effect is mediated via musca-rinic and serotinergic pathway blockade that has been noted to augment deltaand beta activity. Nonetheless, the possibility that antipsychotic medicationhas accounted for a significant proportion of these alterations in the EEGcannot be dismissed in this study. Replication of the present findings in anunmedicated group of subjects with schizophrenia is required.

A decrease in alpha mean frequency was predicted as being in keepingwith this slowing of the power spectrum in schizophrenia (Westphal et al.,1990; Harris et al, 1998; Knott et al., 2001) and secondly, observations of theassociation between mean alpha frequency and cognitive performance (William-son et al., 1989; Gevins, 1998; Klimesch, 1999). Subjects with schizophreniahave been observed to have a loss of cognitive function in both first episode(Bilder et al., 2000) and chronic (Heinrichs & Zakzanis, 1998) schizophreniasuggesting that peak alpha frequency would be decreased in the clinical popu-lations. However, a slowing of alpha mean frequency was observed only inthe group with chronic schizophrenia. This finding was not site specific,although a difference in the pattern of change across the midline sites wasseen, with subjects with schizophrenia increasing their mean alpha frequencyat Pz whereas the control subjects decreased their mean alpha frequency.Symptom pattern is unlikely to explain the difference between the first epi-sode and chronic schizophrenia subjects as increased slow alpha power hasusually been associated with increased negative symptoms (Merrin & Floyd,1992; Sponheim et al., 2000). The two patient groups differed in the levels ofpositive symptoms alone (see Table 1). Further, both clinical groups hadgreater, though not significantly so, alpha power levels than their matchedcontrols, although this does not vitiate the possibility of a relationship be-tween negative symptoms and alpha power. As noted earlier the populationfrom which the subjects with chronic schizophrenia were drawn was likely tobe biased toward greater severity than the first episode schizophrenia group.An alternative explanation for the isolation of this finding to the subjectswith chronic schizophrenia alone could be that slowing of the qEEG mayonly become apparent with chronicity, rather than with treatment non-re-sponse per se.

This study has tested Dynamic Spectral Analysis (DSA), a novel andmathematically rigorous technique in two groups of subjects with schizo-phrenia and compared the finding with those derived from conventional powerspectral analysis. The new technique allows for automated artefact rejectionand capitilizes on the high temporal resolution of the EEG in the frequencydomain. The study has confirmed findings of increased slow wave activity in

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


subjects with schizophrenia, but has also identified a slowing in mean bandfrequency, except in the delta band, in subjects with chronic schizophrenia.Changes in alpha frequency may represent a more stringent marker of dis-ease severity and chronicity than power or amplitude spectrum. This is beinginvestigated by a follow-up study of the subjects with first episode schizo-phrenia from this group.

REFERENCES

Agarwal, R., Gotman, J., Flanagan, D., & Rosenblatt, B. (1998). Automatic EEGanalysis during long-term monitoring in the ICU. Electroencephalography andClinical Neurophysiology, 107, 44–58.

American Psychiatric Association. (1987). Diagnostic and satistical manual of mentaldisorders (3rd ed. Revised). Washington, DC: American Psychiatric Association.

American Psychiatric Association. (1994). Diagnostic and statistical manual of men-tal disorders (4th edition). Washington, DC: American Psychiatric Association.

Barlow, J. S. (1985). Methods of analysis of nonstationary EEGs, with emphasis onsegmentation techniques: A comparative review. Journal of Clinical Neurophysiology,2, 267–304.

Bilder, R. M., Goldman, R. S., Robinson, D., Reiter, G., Bell, L., Bates, J. A., Pappadopulos,E., Wilson, D. F., Alvir, J. M. J., Woerner, M. G., Geisler, S., Kane, J. M., &Lieberman, J. A. (2000). Neuropsychology of first-episode schizophrenia: Initialcharacterization and clinical correlates. American Journal of Psychiatry, 157, 549–559.

Blackman, R. B., & Tukey, J. W. (1959). The measurement of power spectra. NewYork: Dover Publications, Inc.

Blom, J. L., & Anneveldt, M. (1982). An electrode cap tested. Electroencephalographyand Clinical Neurophysiology, 54, 591–594.

Cook, E. W., & Miller, G. A. (1992). Digital filtering: Background and tutorial forpsychophysiologists. Psychophysiology, 29, 350–367.

Creutzfeldt, O.-D., Bodenstein, G., & Barlow, J. S. (1985). Computerized EEG patternclassification by adaptive segmentation and probability density function classifi-cation. Clinical evaluation. Electroencephalography and Clinical Neurophysiology,60, 373–393.

Filon, L. N. G. (1928). On a quadrature formula for trigonometric integrals. Proceed-ings of the Royal Society of Edinburgh, 49, 38–47.

Florian, G., & Pfurtscheller, G. (1995). Dynamic spectral analysis of event-relatedEEG data. Electroencephalography and Clinical Neurophysiology, 95, 393–396.

Gevins, A. (1998). The future of electroencephalography in assessing neurocognitive func-tioning. Electroencephalography and Clinical Neurophysiology, 106, 165–172.

Gratton, G., Coles, M. G., & Donchin, E. (1983). A new method for off-line removal

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


of ocular artifact. Electroencephalography and Clinical Neurophysiology, 55, 468–484.

Guenther, W., Davous, P., Godet, J.-L., Guillibert, E., Breitling, D., & Rondot, P.(1988). Bilateral brain dysfunction during motor activation in type II schizo-phrenia measured by EEG mapping. Biological Psychiatry, 23, 295–311.

Harris C. M. (1998) The Fourier analysis of biological transients. Journal of Neuro-science Methods 83, 15–34.

Harris, A. W. F., Bahramali, H., Lagopoulos, J., Li, W. M., Gordon, E., & Wright, J.J. (1998). The effect of paradigm on alpha frequency in schizophrenia. Interna-tional Journal of Neuroscience, 96, 63–72.

Harris, A. W. F., Gordon, E., Anderson, J., Ritchie, G., McLachlan, C., & Meares, R.(1997). Change in quantified electroencephalography (QEEG) with medicationand altered clinical state in the same subjects with schizophrenia. SchizophreniaResearch, 23, 87–89.

Heinrichs, R. W., & Zakzanis, K. K. (1998). Neurocognitive deficit in schizophrenia:A quantitative review of the evidence. Neuropsychology, 12, 426–445.

Itil, T. M. (1977). Qualitative and quantitative EEG findings in schizophrenia. SchizophreniaBulletin, 3, 61–79.

John, E. R., Prichep, L. S., Alper, K. R., Mas, F. G., Cancro, R., Easton, P., & Sverlov,L. (1994). Quantitative electrophysiological characteristics and subtyping of schizo-phrenia. Biological Psychiatry, 36, 801–826.

Kahn, E. M., Weiner, R. D., Coppola, R., Kudler, H. S., & Schutz, K. (1993). Spectraland topographical analysis of EEG in schizophrenic patients. Biological Psy-chiatry, 33, 284–290.

Karson, C. N., Coppola, R., Morihisa, J. M., & Weinberger, D. R. (1987). Computeredelectroencephalographic activity mapping in schizophrenia—The resting state re-considered. Archives of General Psychiatry, 44, 514–517.

Karson, C. N., Coppola, R., Daniel, D. G., & Weinberger, D. R. (1988). Computer-ised EEG in schizophrenia. Schizophrenia Bulletin, 14, 193–197.

Kay, S. R., Opler, L. A., & Fiszbein, A. (1986a). Positive and Negative SyndromeScale (PANSS). North Tonawanda, NY: Multi-Health Systems Inc.

Keppel, G. (1982). Design and Analysis. A Researcher’s Handbook (2nd ed.). EnglewoodCliffs: Prentice Hall.

Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memoryperformance: A review and analysis. Brain Research Reviews, 29, 169–195.

Klimesch W. (1996). Doppelmayr M, Schimke H, & Pachinger T. Alpha frequency,reaction time and the speed of processing information. Journal of Clinical Neu-rophysiology, 13, 511–518.

Klimesch, W., Schimke, H., & Pfurtscheller, G. (1993). Alpha frequency, cognitiveload and memory performance. Brain Topography, 6, 241–251.

Knott, V., Harr, A., & Ilivitsky, V. (1997). EEG correlates of acute nicotinic andmuscarinic cholinergic blockade: Seperate and combined administration of sco-

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


polamine and mecamylamine in normal human subjects. Human Psychopharma-cology, 12, 573–582.

Knott, V., Labelle, A., Jones, B., & Mahoney, C. (2001). Quantitative EEG in schizophreniaand in response to acute and chronic clozapine treatment. Schizophrenia Research,50, 41–53.

Lambert, T. (1998). Lundbeck Utility for Neuroleptic Dose Conversion. Melbourne:The Multimedia in Psychiatry Unit, University of Melbourne.

Lambert, T. (1999): Switching to Risperdal. Multimedia in Psychiatry Unit. Univer-sity of Melbourne. Melbourne.

Mäkinen, S. (1982). New algorithm for the calculation of the Fourier transform ofdiscrete signals. Review of Scientific Instruments, 53, 627–630.

Melkonian D. (1987). Transients in Neuronal Systems (in Russian). Yerevan: Arme-nian Academy of Sciences.

Melkonian, D., Gordon, E., Rennie, C., & Bahramali, H. (1998). Dynamic spectralanalysis of event-related potentials. Electroencephalography and Clinical Neu-rophysiology, 108, 251–259.

Melkonian, D., Blumenthal T. D., & Meares, R. (2003). High resolution fragmentarydecomposition—A model based method of non-stationary electrophysiologicalsignal analysis. Journal of Neuroscience Methods, 131, 149–159.

Merrin, E. L., & Floyd, T. C. (1992). Negative symptoms and EEG alpha activity inschizophrenic patients. Schizophrenia Research, 8, 11–20.

Mukundan, C. R. (1986). Computed EEG in schizophrenics. Biological Psychiatry,21, 1221–1225.

Niedermeyer, E. (1999). Maturation of the EEG: Development of waking and sleeppatterns. In E. Niedermeyer and F. Lopes da Silva (Eds.), Electroencephalography(4th ed., pp. 189–214). Philadelphia: Lippincott, Williams & Wilkins.

Nuwer, M. R. (1988). Quantitative EEG: I. Techniques and problems of frequencyanalysis and topographic mapping. Journal of Clinical Neurophysiology, 5,1–43.

Pivik, R. T., Broughton, R. J., Coppola, R., Davidson, R. J., Fox, N., & Nuwer, M.R. (1993). Guidelines for the recording and quantitative analysis of electroencephalo-graphic activity in research context. (Committee report). Psychophysiology, 30,547–558.

Roitbak, A. I., Fanardjian, V. V., Melkonyan D. S., & Melkonyan A. A. (1987).Contribution of glia and neurons to the surface-negative potentials of the cere-bral cortex during its electrical stimulation. Neuroscience, 20, 1057–1067.

Schütte, J. (1981). New fast Fourier transform algorithm for linear system analysisapplied in molecular beam relaxation spectroscopy. Review of Scientific Instru-ments, 52, 400–404.

Sponheim, S. R., Clementz, B. A., Iacono, W. G., & Beiser, M. (1994). Resting EEGin first-episode and chronic schizophrenia. Psychophysiology, 31, 37–43.

Sponheim, S. R., Clementz, B. A., Iacono, W. G., & Beiser, M. (2000). Clinical and

Q5

Q5

Q6

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.


biological concomitants of resting state EEG power abnormalities in schizophre-nia. Biological Psychiatry, 48, 1088–1097.

SPSS. (1999). Statistical Package for the Social Sciences. v. 10.0.5. Apache SoftwareFoundation.

Stevens, J. R., & Livermore, A. (1982). Telemetered EEG in schizophrenia: Spectralanalysis during abnormal behaviour episodes. Journal of Neurology, Neurosur-gery and Psychiatry, 45, 385–395.

Westphal, K. P., Grozinger, B., Diekmann, V., Scherb, W., Rees, J., Leibing, U., &Kornhuber, H. H. (1990). Slower theta activity over the mid-frontal cortex inschizophrenic patients. Acta Psychiatrica Scandinavica, 81, 132–138.

Williamson, P. C., Kutcher, S. P., Cooper, P. W., Snow, W. G., Szalai, J. P., Kaye,H., Morrison, S. L., Willinsky, R. A., & Mamelak, M. (1989). Psychological,topographical EEG, and CT scan correlates of frontal lobe function in schizo-phrenia. Psychiatry Research, 29, 137–149.

World Health Organization. (1992). Composite International Diagnostic Interview.Geneva: WHO.

Int J

Neu

rosc

i Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

CD

L-U

C R

iver

side

on

11/0

2/14

For

pers

onal

use

onl

y.

dynamic spectral analysis findings in first episode and chronic schizophrenia

Documents