Epilepsy Research (2008) 82, 29—37

journa l homepage: www.e lsev ier .com/ locate /ep i lepsyres

Volume determination of amygdala and hippocampusat 1.5 and 3.0 T MRI in temporal lobe epilepsy

Jasmin E. Scorzina,∗,1, Sabine Kaadena,1, Carlos M. Quesadab,Christian-Andreas Müllera, Rolf Fimmersc, Horst Urbachd,Johannes Schramma

a Department of Neurosurgery, University of Bonn, Sigmund-Freud-Str. 25, 53105 Bonn, Germanyb Department of Epileptology, University of Bonn, Germanyc Institute of Medical Biometry, Informatics and Epidemiology, University of Bonn, Germany

d Department of Radiology and Neuroradiology, University of Bonn, Germany

Received 10 January 2008; receivedAvailable online 8 August 2008

KEYWORDSVolumetry;VBM;Temporal lobeepilepsy;3D MRI;3.0 T

∗ Corresponding author. Tel.: +49 22fax: +49 228 287 14758.

E-mail addresses: Jasmin.Scorzin@Sabine.Kaaden@ukb.uni-bonn.de (J.E

1 These authors contributed equally

0920-1211/$ — see front matter © 20doi:10.1016/j.eplepsyres.2008.06.012

in revised form 3 June 2008; accepted 29 June 2008

Summary Since magnetic resonance imaging (MRI) technique is constantly evolving withhigher field strength scanners, the question arises whether images from different field strengthscanners can be used interchangeably for scientific and clinical purposes. We address this issuein a study group of patients with temporal lobe epilepsy (TLE).

Two different quantification methods for analysing structural (MRI) were used. Conventionalvolumetry was performed by manually tracing amygdala and hippocampus volumes on both1.5 and 3 T scans of 10 TLE patients. Additionally a voxel-based morphometry (VBM)-basedextraction of those structures was conducted.

As an answer to the main question, it was determined that the volumetrically derived volumesof amygdala and hippocampus from 1.5 and 3.0 T images did not differ. Our findings concerningthe volumetry are consistent with findings in healthy controls, thus offering the possibilityto use volumetry of the different scanners interchangeably. The results of the VBM-analyses

show satisfying inter-scanner volume quantification but not consistent enough to be deemedinterchangeable. Further investigations analysing the outcomes of conventional VBM of differentfield strength scanners are nece© 2008 Elsevier B.V. All rights re

8 287 16521;

ukb.uni-bonn.de,. Scorzin).to this work.

I

Arcee

08 Elsevier B.V. All rights reserved.

ssary.served.

ntroduction

frequent diagnosis in medically intractable mesial tempo-al lobe epilepsy (TLE) is hippocampal sclerosis, and theseases are often treated with resective epilepsy surgery. Thextent of damage in the affected region as well as thextent of surgical resection are considered critical determi-

3

nutpsot

aTfittTis

smta2Ribaai

tcsr

M

P

Tifccsuatemepdct(

M

3ts

T

2rsa

av2

M

AfBDafiqq

pth(K2ttdo

wasstp

M

VrTas

tr

ih(tTim(

V

TI

0

ants of cognitive and seizure outcome. Many researchersse magnetic resonance imaging (MRI) images to quantifyhe volumes of temporomesial structures, namely the hip-ocampus (HC) and amygdala (AM). The small size of thesetructures and the adjacent position to each other and tother brain structures demand high resolution and visualisa-ion on medical imaging.

Magnetic resonance imaging is the standard technique tonalyze and detect structural changes of the HC and AM inLE. Technological progress with the introduction of highereld strength systems in clinical settings is occurring, sohat 3.0 T scanners are nowadays available in most hospi-als, especially if they offer an epilepsy surgery program.his hardware change has implications for longitudinal stud-

es, since it has always been said that images from differentcanners are not easily compared.

The aim of this study is to compare MRI at different fieldtrengths (1.5 T versus 3.0 T) in matters of AM and HC byanually outlining these regions of interest (ROIs), a tradi-

ional volumetry approach. Additionally, a second proceduredapted from VBM (Ashburner and Friston, 2000; Good et al.,001; Mechelli et al., 2005) was used to compare the sameOIs with both field strengths. Contrary to the former, VBM

s mainly computer-based therefore faster and less affectedy the investigator. Since its first development by Ashburnernd Friston (2000) it has been applied to many studies onlmost every neurological and psychiatric illness, as well asn healthy volunteers studies.

This study served not the purpose to directly comparehe outcomes of volumetric versus a VBM approach. For thatomparison a ‘‘gold standard’’ would be needed, which iso far not available. In the present paper we will thereforeeport the findings of volumetry and VBM separately.

aterials and methods

atients

en patients (six female) with the clinical diagnosis of medicallyntractable mesial temporal lobe epilepsy (mTLE) were collectedrom our epilepsy surgery program. Approval from the local ethicsommittee was obtained, and all patients gave written informedonsent before their inclusion. The mean age (at the time ofurgery) was 36.5 years and ranged from 18 to 52 years. All patientsnderwent MRI on both scanners before epilepsy surgery (7 selectivemygdalohippocampectomies, 2 standard 2/3 temporal lobe resec-ions and 1 mesial lesionectomy). The mean time between the twoxaminations was 5.6 months (range 0.3—24 months; median 2.5onths). Surgical intervention was not delayed by the double MRI

xamination. Preoperative radiological diagnosis showed hippocam-al sclerosis in eight cases, one case of ganglioglioma (which was aysembryoplastic neuroepithelial tumor after final histopathologi-al examination) and one case of dysplasia. The affected side washe right hemisphere in six patients, and the left on the other fourno bilateral cases).

RI protocol

D imaging studies were performed preoperatively at two MR sys-ems (Philips Gyroscan, Eindhoven, NL), differing in their fieldtrengths (1.5 and 3.0 T).

The protocol for the 1.5 T scan was a sagittally acquired 3D1-weighted gradient echo sequence with 1 mm3 isotropic voxel,

taood

J.E. Scorzin et al.

56-mm field of view (FOV), 84% rectangular field of view (RFOV),epetition time (TR) of 15 ms, echo time (TE) of 3.6 ms, number ofignal averages (NSA) = 1, 256 × 256 matrix, 140 slices and 30◦ flipngle.

The 3.0 T scans were also acquired in sagittal orientation. It was3D T1-weighted turbo field echo sequence with 1 mm3 isotropic

oxel, 256-mm FOV, 95% RFOV, TR of 8.2 ms, TE of 3.8 ms, NSA = 1,56 × 256 matrix, 140 slices and 8◦ flip angle.

RI analysis and processing

ll image processing steps and volume measurements were per-ormed using ANALYZETM 3.1 software package of the Mayoiomedical Imaging Resource Rochester, MN, USA (Robb, 2001). RawICOM format data were first transformed to ANALYZE format fornalysis. The data sets underwent intensity normalization, spatialltering and correction for field inhomogeneities due to radiofre-uency non-uniformity. These corrections are assumed to benefituantitative measurements (Sled et al., 1998).

To align coronal slices perpendicular to the long axis of the hip-ocampal formation, which is widely accepted by many researcherso reduce volume averaging and to obtain superior visualization withigh efficiency and reliability in respect to volumetric measurementBartzokis et al., 1993, 1998; Cendes et al., 1993; Jack et al., 1995;im et al., 1995; Laakso et al., 1997; Pantel, 1998; Pantel et al.,000), a reference line (chiasmatico-commissural line) is defined inhe midsagittal pilot and the coronal slices are tilted perpendicularo this plane afterwards. Nevertheless postscan-processing proce-ures are necessary to achieve exact and reliable adjustment in allrientations.

Aligning the coronal slices perpendicular to the HC length axis,hich is rather parallel to the lateral fissure, was achieved through3D registration tool. The chiasmatico-commissural line (CH-PC) is

ituated at the midbrain—diencephalic junction in the midsagittallice and is defined as a line tangential to the superior border ofhe optic chiasm (CH) anteriorly and to the inferior border of theosterior commissure (PC) posteriorly.

anual volume measurement

olume measurements of the HC and AM of both hemispheres,espectively were accomplished by mouse-driven manual tracing.he ROIs were outlined on each slice. The volumes were calculatedutomatically by pixel counting and adding up the single coronallice areas (see Fig. 1).

The anatomical landmarks and boundaries used follow basicallyhose described by the Duvernoy Atlas (Duvernoy, 1988, 1998) andecent guidelines of Pantel (1998) and Pantel et al. (2000).

Measuring isolated hippocampal grey matter rather than assess-ng combined white and grey matter is considered to achieve aigher sensitivity and accuracy in detecting volume differencesSheline et al., 1996) since grey matter lesions mainly consti-ute classical hippocampal sclerosis for example (Duvernoy, 1998).herefore, alveus and most of efferent pathway of the HC includ-

ng the fimbria and the fornix were excluded from hippocampaleasurement. For detailed tracing borders see recent publication

Mueller et al., 2007).

BM-procedure

he same T1-weighted images used for the volumetry were used.nstead of the usual whole-brain voxel-wise analysis, some addi-

ional tools for extracting volume values from the four ROIs werepplied. Analysis was performed using SPM5 (www.fil.ion.ucl.ac.uk)n MATLAB 6.5 for Windows (www.mathworks.com), following theptimized VBM procedure (Good et al., 2001). All images were ren-ered to a common space (MNI-template), and then segmented into

Temporal lobe epilepsy 31

Figure 1 Outlined hippocampus on coronal slices at 1.5 and 3.0 T images on levels of hippocampal head (a), body (b) and tail(c). (a) Hippocampal head (HC), amygdala (AM), temporal horn of the lateral ventricle (TH), parahippocampal gyrus (PHG), (b)

gyrurn o

Tar

t

vmsafVa

csiaat

hippocampal body (HC), temporal stem (TS), parahippocampaltemporal stem (TS), parahippocampal gyrus (PHG), temporal hocistern (QC).

only grey-substance images. As our study concentrates on volumesof two structures (AM and HC) which are grey matter only structures,the analysis was performed only on grey matter images. The greymatter images were corrected for the small local volume changesperformed during the non-linear normalisation, a process calledmodulation. By applying this modulation, images are encoded withvolume information, as has been described before (Good et al.,2001; Haier et al., 2004; Keller et al., 2004). Optional modulationof non-linear effects only Modulation was applied only for the non-linear effect, as described by Christian Gaser from the University ofJena, Germany (http://dbm.neuro.uni-jena.de), making correctingfor total brain volume of the individual unnecessary.

The volumes of interest (VOIs) for the respective temporome-sial structures were drawn in a high-resolution 3D-T1-image, whichwas already aligned to the same common space which had beenused to normalize the images from the 10 subjects. One rater(S.K.), experienced in manual volumetry, performed the draw-ing of the masks following the above-described protocol used forthe volumetric analyses using the free-licensed MRIcro software(www.sph.sc.edu/comd/rorden/mricro.html). Examples of the ROIs

are displayed in Fig. 2.

The same manually obtained ROIs were obtained on the identicalhigh-resolution image using an atlas developed for SPM from theWake-Forest-University (WFU-pick-Atlas Version 2.4) (Lancaster etal., 2000; Maldjian et al., 2003) (http://www.fmri.wfubmc.edu).

eaFtv

s (PHG), ambient cistern (AC), and (c) hippocampal tail (HC),f the lateral ventricle (TH), crus of fornix (CF), quadrigeminal

his secondary procedure was chosen to test whether our manuallynd automatically drawn WFU-atlas masks would generate the sameesults.

In brief, these masks represent the brain structures which willhen be automatically extracted and serve as so-called VOIs.

Once the VOIs had been drawn, the mean signal intensity of alloxels included in every VOI were calculated by using another auto-atic SPM tool (http://marsbar.sourceforge.net/). This makes our

tudy no longer voxel-based but VOI-based. The benefit of aver-ging all voxel from a whole ROI is that we gain statistical poweror making inter-subject analysis, which is the case concerning theBM procedure. Those values were extracted from both manuallynd automatically obtained VOIs for posterior analysis.

We would like to point out, that the VBM-derived approacharries unresolved problems in terms of comparing different fieldtrengths. 3.0 T is supposed to come along with a better image qual-ty, to provide a better temporal, spatial and spectral resolution, butlso a higher susceptibility for artefacts, resulting in geometric shiftnd signal loss (Filippini et al., 2006; Yang et al., 2004). We supposehat all these parameters may influence the VBM procedure; how-

ver this will not be discussed further in detail in the present studynd should be investigated in systematically conducted studies.urthermore, we may not control the above-mentioned parame-ers and this was not our purpose, because we focused mainly onolumetry. Thus, one cannot draw conclusions from our analyses

32J.E.

Scorzinet

al.

Figure 2 Examples of manually drawn masks of amygdala and hippocampus on coronal slices. Masks were drawn into a standard brain (MNI-template). AM = left amygdala;HC = left hippocampus.

Temporal lobe epilepsy 33

Table 1 R2 for possible confounding variables

R2 (1.5 T vs. 3.0 T) Total Hemisphere Affected side Gender

Left (n = 10) Right (n = 10) Left (n = 8) Right (n = 12) Female (n = 12) Male (n = 8)

AM 0.778 0.736 0.856 0.803 0.757 0.780 0.794

isphe

pr

Bmo

aimm

VFfrc

R

V

Wsfifbt

HC 0.953 0.945 0.962

Different R2 (coefficient of determination) for the subgroups hem

described below, but we decided to report this preliminary trial forthe interest of the reader.

Data and statistical analysis

VolumetrySPSS (14.0) software was used for statistical analysis. For manualvolumetry each data set (n = 20, 10 sets at 1.5 T, 10 sets at 3.0 T)with four ROIs, respectively (HC and AM of both hemispheres) wasmeasured by the same experienced rater (J.E.S.) in random order.A second measurement of all volumes was performed again by thesame rater in random order with a 1-month time interval betweenthe two ratings.

Intrarater reliability was calculated for AM and HC separatelyusing intraclass-correlation-coefficients (ICC).

Two-tailed Pearson’s correlation coefficient (r) was calculatedto describe the degree of linear relationship between the two mea-surements.

Measurement variability (variability between the first and thesecond measurement of the same data set) was analyzed separatelyfor 1.5 and 3.0 T. We looked for systematic differences in the valuesbetween the first and second measurement.

Interscanner variability estimates the differences of the meansof the repeated volume measurements at different field strengths.R2 (coefficient of determination) was used to show the proportionof variance that the values have in common. The overall correlationcoefficient was specified to detect possible affection by subgroups(gender, hemisphere and affected side); the results of R2 are listedin table form (Table 1). Additionally bivariate partial correlationswere performed.

A known problem in comparing the two methods is, that althoughtwo measurements are highly correlated, the means might be dif-ferent (e.g. one method measures systematically lower or highervalues), resulting in significant differences of the mean. Thatimplies, that correlations alone are not the appropriate statistical

cafita

Figure 3 Relationship between the repeated measurements of am

0.966 0.943 0.958 0.945

re, gender and affected side.

rocedure to investigate whether to methods generate comparableesults.

As proposed by Bland and Altman (1986), we use thereforeland—Altman plots showing the differences and means of the twoethods thus permitting a descriptive evaluation of the agreement

f two measurements.To describe the agreement between the compared values (1.5

nd 3.0 T) in each case the difference of two paired measurementss plotted against the mean of the two measurements. It is recom-ended that 95% of the data points should lie within ±2 S.D. of theean difference (limits of agreement).

OI-procedurerom the described procedures we received mere numeric valuesor the 1.5 and 3.0 T images, which represent the volumes of theespective VOIs. These were analyzed using two-tailed Pearson’sorrelation coefficient and linear regression coefficients (R2).

esults

olumetry

e found high linear relationship between the first andecond measurement. Two-tailed Pearson’s correlation coef-cient (r) was 0.994 (p = 0.01) for AM and 0.997 (p = 0.01)or HC. Scatter plots show a large distribution of values foroth regions (Fig. 3), which was expected since atrophy ofhe affected hemisphere produce smaller values in volume

ompared to the healthy, unaffected hemisphere. There wasslight systematical difference in the values between the

rst and second measurement. Fig. 4 shows the distribu-ion of the values’ differences around zero when plottedgainst their means. The median of the measurements’ dif-

ygdala (AM) and hippocampus (HC) shown in separate plots.

34 J.E. Scorzin et al.

F asurev

fimH(

tt(sbsd9ua

tbO

Tsw2iacswo(

V

Fa±

igure 4 Bland—Altman plots show the distribution of the mealues in the second measurement.

erences is 17.2 mm3 for HC and 11.58 mm3 for AM, whichndicates a tendency towards smaller values in the secondeasurement. Intrarater reliability was 0.997 for repeatedC measurement and 0.993 for repeated AM measurement

single values).For inter-scanner variability the differences between

he measurements at 1.5 and 3.0 T where plotted againstheir means (Fig. 5). We took mean values of the tworepeated) measurements into account. Two-tailed Pear-on’s correlation coefficient (r) showed high correlationetween the different field strengths for hippocampal mea-urement (r = 0.975; R2 = 0.952). The high coefficient ofetermination shows that the values at 1.5 and 3.0 T have5% of their variance in common. The agreement of the val-es was good as well. The limit of agreement was 306.3 mm3

3

bove to 231.4 mm below the mean difference.For the AM measurements, two-tailed Pearson’s correla-

ion coefficient (r) showed a slightly decreased correlationetween the different field strengths (r = 0.882; R2 = 0.778).nly about 78% of the values variance is in common.

Il1tc

igure 5 Bland—Altman plots show the distribution of the measurere separated for amygdala and hippocampus. The lines indicate the2 S.D. for the hippocampus measurements, whereas only 90% of da

ments’ differences. There is a slight tendency towards smaller

he Bland—Altman plot shows 10% of the data points out-ide the confidence interval. Two data points out of 20ere outside the limits of agreement (259.8 above to58.9 mm3 below the mean difference). Possibly confound-ng variables like gender, affected side (side of disease)nd hemisphere had no significant effect on the overallorrelations. Bivariate partial correlation test showed noignificant correlations between these subvariables eachith the different field strengths. The different coefficientsf determination for the subgroups are given in table formTable 1).

OI analyses

n the following paragraph the Pearson’s correlation andinear regression coefficients of the VOI analyses for the.5 and 3.0 T measurements are provided, starting withhe manually drawn masks. Two-tailed Pearson’s correlationoefficient (r) was 0.766** (p = 0.010) for left AM and 0.901**

ments’ differences between different field strengths. The plotsmean difference and ±2 S.D. 95% of data points lie within the

ta points lie within ±2 S.D. for the amygdala measurements.

Temporal lobe epilepsy 35

theR2: le

cvodm

Figure 6 Scatterplots showing the linear relationship (R2) forand manually drawn masks. Manually drawn masks (MD): linearR2: left AM = 0.598; right AM = 0.791.

(p < 0.001) for right AM, 0.497 (p = 0.144) for left HC and forright HC the correlation coefficient was 0.816** (p = 0.004).The Pearson correlations for the WFU-atlas masks were0.773** (p = 0.009) for left, 0.889** (p = 0.001) for right AM,0.422 (p = 0.224) for left and 0.719* (p = 0.019) for rightHC. Correlations for automatically segmented grey and

white matter and cerebrospinal fluid were situated in equalor even lower level (GM: r = 0.579; WM: r = 0.944**; CSF:r = 0.608). The Pearson’s correlation for total brain volumewas 0.843** (p = 0.002).

vmv

Figure 7 Scatterplots showing the linear relationship (R2) for theatlas and manually drawn masks. Manually drawn masks (MD): linearlinear R2: left HC = 0.178; right HC = 0.517.

VOI-segmented amygdala volumes at 1.5 and 3 T for WFU-atlasft AM = 0.586; right AM = 0.812. WFU-atlas masks (WFU): linear

The scatterplots also presenting the regression-oefficients for the manually and WFU-atlas masks-derivedolumes are shown in Figs. 6 and 7. Between 24.7 and 81.2%f the variance can be explained concerning the manuallyrawn masks, whereas the r2 values for the WFU-atlasasks lie between 17.8 and 79.1%.

Fig. 8 shows the regression between 1.5 and 3.0 T derived

olumes separately for manually drawn versus WFU-atlasasks VOIs and one can see, that both methods come to

ery similar results.

VOI-segmented hippocampus volumes at 1.5 and 3 T for WFU-R2: left HC = 0.247; right HC = 0.666. WFU-atlas masks (WFU):

36 J.E. Scorzin et al.

e 1.5

D

TumgfieaasasTbdmdidi

o(qima

tfibwtat(rf

bmiffa

mfsia

(

(

hobchm

Figure 8 Scatterplots and R2 for th

iscussion

here was no systematic difference between HC and AM vol-me measurements performed at 1.5 and at 3.0 T. For HCeasurements we found high correlation (r = 0.975) and a

ood agreement of the values. This finding conforms withndings of other authors (Bartzokis et al., 1993; Briellmannt al., 2001). Both found good interscanner reliability,lthough one measured different brain structures at 1.5 Tnd at 0.5 T in five subjects (Bartzokis et al., 1993). Theecond author measured the HC of eight healthy controlst 1.5 T and at 3.0 T (Briellmann et al., 2001). He found noystematic difference between HC volume measurements.hey postulate that the manual tracing technique mighte the source of measurement variability rather than theifferent field strength. None of the authors analyzed theeasurement of a rather small object like the AM as weid. The measurement of the AM in our study showed goodntrarater reliability (ICC = 0.993), but we found a slightlyecreased correlation and not so good agreement in thenterscanner measurement.

This observation might be due to the small size of thebject. The AM is known to be not easily traced on MRIPantel et al., 2000) and even slight differences in imageuality and contrast might significantly influence the visual-zation and demarcation of a small object, especially whenorphological changes like sclerosis or other pathologies

ffect the normal shape and volume of the object.Nevertheless, there are also studies in which a correla-

ion between more accurate delineation of structures andeld strength and improvement of measurement repeata-ility (Levy-Reis et al., 2000; Wieshmann et al., 1998)as observed. Furthermore at higher field strength (4.0 T)

he volumes were found to be systematically smaller. The

uthors concluded that higher resolution at 4.0 T comparedo 1.5 T provides more structural information. As othersBartzokis et al., 1993; Briellmann et al., 2001) we could noteproduce this systematic interscanner difference, albeitor the AM a decreased correlation and poorer agreement

efi

ui

T vs. 3.0 T VOI-segmented volumes.

etween values at 1.5 T and at 3.0 T was obtained. Thiseans that imaging at different field strength — unless the

maging protocol and sequence do not differ too much at dif-erent field strengths — is applicable in volumetry at leastor bigger temporomesial structures like the HC interchange-bly and without systematic volume differences.

Concerning the VOI analyses, we were interested in twoain questions: (1) do the automatically derived volumes

or HC and AM differ between the different field-strengthcanners; and (2) does implementing a manually drawn maskmprove the results compared to using the masks defined onstandard atlas.

1) In accordance with the results concerning the volume-try, apart from we obtained satisfactory correlations andregression coefficients for the 1.5 and 3.0 T segmentedvolumes resulting from the VOI analyses. For the leftHC, apparently lower correlation and regression coeffi-cients were achieved, irrespective of the method thatwas used (WFU versus manually drawn). However, thereason for this remains unclear.

2) We observed no meaningful differences using the man-ual masks or WFU-atlas masks that would justify thetime and effort creating own masks, instead of eas-ily using the ones provided by the WFU-Pick-Atlas(Lancaster et al., 2000; Maldjian et al., 2003).

Apart from the restrictions we mentioned before, weave to keep in mind that these analyses were conductedn images from epilepsy patients, with the pathology possi-ly influencing the outcome. The analyses should be alsoonducted for a healthy sample, which might result inigher correlations and linear relationships. However, asentioned before, one study, which volumetrically analyzed

ight healthy subjects found no differences due to differenteld strength (Briellmann et al., 2001).

Using the VOI-based approach as a supplement to the vol-metric analyses, we get the results for one method, whichs generally seen as influenced by the raters who perform the

H

J

K

K

L

L

L

M

M

M

P

P

R

S

S

W

Temporal lobe epilepsy

measurements and on the other hand, for a more objectivemethod.

In various studies, volumetry has proven its anatomicalvalidity, whereas for VBM it is still under current discussionhow reliable the results are and what they really reflectand measure. Our findings concerning the volumetry at 1.5and 3.0 T are consistent with those found in healthy personsby Briellmann et al. (2001) and the results based on thevbm-derived approach appear promising. However, it is ofclinical interest and should be investigated in further studieswhether reliable results will be obtained using voxel-basedmorphometry at different field strengths.

Acknowledgements

This study was part of the SFB-TR3 collaborative researchproject ‘Mesial Temporal Lobe Epilepsies’ supported by agrant of the Deutsche Forschungsgemeinschaft (DFG). CMQuesada is supported by the BMBF (Grant 01GW0511). Spe-cial thanks go to Devin K. Binder for his editing comments(University of California, Irvine). We thank Prof. Elger, Prof.Helmstaedter and colleagues of the Department of Epilep-tology, University of Bonn, for evaluating patients and forcollaboration.

References

Ashburner, J., Friston, K.J., 2000. Voxel-based morphometry—–themethods. Neuroimage 11 (6 Pt 1), 805—821.

Bartzokis, G., Mintz, J., Marx, P., Osborn, D., Gutkind, D., Chiang,F., Phelan, C.K., Marder, S.R., 1993. Reliability of in vivo volumemeasures of hippocampus and other brain structures using MRI.Magn. Reson. Imag. 11 (7), 993—1006.

Bartzokis, G., Altshuler, L.L., Greider, T., Curran, J., Keen, B.,Dixon, W.J., 1998. Reliability of medial temporal lobe volumemeasurements using reformatted 3D images. Psychiatr. Res. 82(1), 11—24.

Bland, J.M., Altman, D.G., 1986. Statistical methods for assess-ing agreement between two methods of clinical measurement.Lancet 1 (8476), 307—310.

Briellmann, R.S., Syngeniotis, A., Jackson, G.D., 2001. Comparisonof hippocampal volumetry at 1.5 T and at 3 T. Epilepsia 42 (8),1021—1024.

Cendes, F., Andermann, F., Gloor, P., Evans, A., Jones-Gotman, M.,Watson, C., Melanson, D., Olivier, A., Peters, T., Lopes-Cendes,I., et al., 1993. MRI volumetric measurement of amygdalaand hippocampus in temporal lobe epilepsy. Neurology 43 (4),719—725.

Duvernoy, H., 1988. The human hippocampus. In: An Atlas of AppliedAnatomy. JFBergmann, Munich.

Duvernoy, H., 1998. The Human Hippocampus: Functional Anatomy,Vascularization and Serial Sections with MRI. Springer, Berlin,Heidelberg.

Filippini, N., Scassellati, C., Boccardi, M., Pievani, M., Testa, C.,Bocchio-Chiavetto, L., Frisoni, G.B., Gennarelli, M., 2006. Influ-

ence of serotonin receptor 2A His452Tyr polymorphism on braintemporal structures: a volumetric MR study. Eur. J. Hum. Genet.14 (4), 443—449.

Good, C.D., Johnsrude, I.S., Ashburner, J., Henson, R.N., Friston,K.J., Frackowiak, R.S., 2001. A voxel-based morphometric study

Y

37

of ageing in 465 normal adult human brains. Neuroimage 14 (1Pt 1), 21—36.

aier, R.J., Jung, R.E., Yeo, R.A., Head, K., Alkire, M.T., 2004.Structural brain variation and general intelligence. Neuroimage23 (1), 425—433.

ack Jr., C.R., Theodore, W.H., Cook, M., McCarthy, G., 1995.MRI-based hippocampal volumetrics: data acquisition, normalranges, and optimal protocol. Magn. Reson. Imag. 13 (8),1057—1064.

eller, S.S., Wilke, M., Wieshmann, U.C., Sluming, V.A., Roberts, N.,2004. Comparison of standard and optimized voxel-based mor-phometry for analysis of brain changes associated with temporallobe epilepsy. Neuroimage 23 (3), 860—868.

im, J.H., Tien, R.D., Felsberg, G.J., Osumi, A.K., Lee, N., Fried-man, A.H., 1995. Fast spin-echo MR in hippocampal sclerosis:correlation with pathology and surgery. AJNR Am. J. Neuroradiol.16 (4), 627—636.

aakso, M.P., Juottonen, K., Partanen, K., Vainio, P., Soininen, H.,1997. MRI volumetry of the hippocampus: the effect of slicethickness on volume formation. Magn. Reson. Imag. 15 (2),263—265.

ancaster, J.L., Woldorff, M.G., Parsons, L.M., Liotti, M., Freitas,C.S., Rainey, L., Kochunov, P.V., Nickerson, D., Mikiten, S.A.,Fox, P.T., 2000. Automated Talairach atlas labels for functionalbrain mapping. Hum. Brain Mapp. 10 (3), 120—131.

evy-Reis, I., Gonzales-Atavales, J.B., King, D.S., French, J.A.,Alsop, D.C., Detre, J.A., 2000. Hippocampal imaging and vol-umetry in temporal lobe epilepsy at 1.5 and 4.0 T. Proc. Int.Soc., Magn. Reson. Med. 8, 15.

aldjian, J.A., Laurienti, P.J., Kraft, R.A., Burdette, J.H., 2003.An automated method for neuroanatomic and cytoarchitectonicatlas-based interrogation of fMRI data sets. Neuroimage 19 (3),1233—1239.

echelli, A., Price, C.J., Friston, K.J., Ashburner, J., 2005.Voxel-based morphometry of the human brain: methods andapplications. Curr. Med. Imag. Rev. 1 (2), 105—113.

ueller, C.A., Scorzin, J., Koenig, R., Urbach, H., Fimmers, R.,Zentner, J., Lehmann, T.N., Schramm, J., 2007. Comparisonof manual tracing versus a semiautomatic radial measurementmethod in temporal lobe MRI volumetry for pharmacoresistantepilepsy. Neuroradiology 49, 189—201.

antel, J. 1998. Hippocampus Tracing Guidelines, fromhttp://moniz.psychiatry.uiowa.edu/.

antel, J., O’Leary, D.S., Cretsinger, K., Bockholt, H.J., Keefe,H., Magnotta, V.A., Andreasen, N.C., 2000. A new method forthe in vivo volumetric measurement of the human hippocam-pus with high neuroanatomical accuracy. Hippocampus 10 (6),752—758.

obb, R.A., 2001. The biomedical imaging resource at Mayo clinic.IEEE Trans. Med. Imag. 20 (9), 854—867.

heline, Y.I., Wang, P.W., Gado, M.H., Csernansky, J.G., Vannier,M.W., 1996. Hippocampal atrophy in recurrent major depression.Proc. Natl. Acad. Sci. USA 93 (9), 3908—3913.

led, J.G., Zijdenbos, A.P., Evans, A.C., 1998. A nonparametricmethod for automatic correction of intensity nonuniformity inMRI data. IEEE Trans. Med. Imag. 17 (1), 87—97.

ieshmann, U.C., Free, S.L., Stevens, J.M., Shorvon, S.D., 1998.Image contrast and hippocampal volumetric measurements.

Magn. Reson. Imag. 16 (1), 13—17.

ang, P.C., Nguyen, P., Shimakawa, A., Brittain, J., Pauly, J.,Nishimura, D., Hu, B., McConnell, M., 2004. Spiral magnetic res-onance coronary angiography—–direct comparison of 1.5 T vs. 3 T.J. Cardiovasc. Magn. Reson. 6 (4), 877—884.