Effectiveness of four different clinical fMRI paradigms for preoperative regional determination of language lateralization in patients with brain tumors

FUNCTIONAL NEURORADIOLOGY

Effectiveness of four different clinical fMRI paradigmsfor preoperative regional determination of languagelateralization in patients with brain tumors

Domenico Zacà & Joshua P. Nickerson & Gerard Deib &

Jay J Pillai

Received: 31 December 2011 /Accepted: 12 June 2012 /Published online: 29 June 2012# Springer-Verlag 2012

AbstractIntroduction Blood oxygen level-dependent functionalmagnetic resonance imaging (fMRI) has demonstrated itscapability to provide comparable results to gold standardintracarotid sodium amobarbital (Wada) testing for preoper-ative determination of language hemispheric dominance.However, thus far, no consensus has been established re-garding which fMRI paradigms are the most effective forthe determination of hemispheric language lateralization inspecific categories of patients and specific regions of interest(ROIs).Methods Forty-one brain tumor patients who performedfour different language tasks—rhyming (R), silent wordgeneration (SWG) sentence completion, and sentence listen-ing comprehension (LC)—for presurgical language map-ping by fMRI were included in this study. A statisticalthreshold-independent lateralization index (LI) was calcu-lated and compared among the paradigms in four differentROIs for language activation: functional Broca’s (BA) andWernicke’s areas (WA) as well as larger anatomically de-fined expressive (EA) and receptive (RA) areas.Results The two expressive paradigms evaluated in thisstudy are very good lateralizing tasks in expressive languageareas; specifically, a significantly higher mean LI value wasnoted for SWG (0.36±0.25) compared to LC (0.16±0.24,

p00.009) and for R (0.40±0.22) compared to LC (0.16±0.24, p00.001) in BA. SWG LI (0.28±0.19) was higherthan LC LI (0.12±0.16, p00.01) also in EA. No significantdifferences in LI were found among these paradigms in WAor RA.Conclusions SWG and R are sufficient for the determina-tion of lateralization in expressive language areas, whereasnew semantic or receptive paradigms need to be designedfor an improved assessment of lateralization in receptivelanguage areas.

Keywords BOLD fMRI . Presurgical mapping . Languagelateralization . Brain tumor

Introduction

Presurgical mapping of language and sensorimotor cortex inpatients with brain tumors currently represents the mostcommonly utilized clinical application of blood oxygenlevel-dependent (BOLD) functional magnetic resonance im-aging (fMRI). Substantial concordance between the resultsof intraoperative cortical stimulation mapping, the electro-physiologic gold standard technique for brain mapping, andpreoperative fMRI has been reported in multiple studies,although variable sensitivities and specificities of fMRI foreloquent cortical localization have been reported acrossthese studies [1–10]. Some of the discordance that has beenreported is likely related to the inherent differences andlimitations of the respective techniques, and these minorlimitations certainly have not precluded widespread clinicalutilization of BOLD fMRI for presurgical planning.

Preoperative assessment of language hemispheric domi-nance heavily influences the surgical management ofpatients with brain lesions in spatial proximity to essentialeloquent language cortices and underlying eloquent white

D. Zacà :G. Deib : J. J. Pillai (*)Division of Neuroradiology, Russell H. Morgan Departmentof Radiology and Radiological Science, The Johns HopkinsUniversity School of Medicine & The Johns Hopkins Hospital,600 N. Wolfe Street, Phipps B-100,Baltimore, MD 21287, USAe-mail: [email protected]

J. P. NickersonDepartment of Radiology, The University of VermontSchool of Medicine/Fletcher Allen Healthcare,111 Colchester Ave.,Burlington, VT 05401, USA

Neuroradiology (2012) 54:1015–1025DOI 10.1007/s00234-012-1056-2

matter tracts. This is true because such knowledge of lan-guage dominance is critical for the determination of theoverall feasibility and safety of lesion resection. Althoughintraoperative electrocortical stimulation mapping is the ac-cepted gold standard for language mapping, it has somemajor limitations in the assessment of language dominancerelated to the limited craniotomy exposure available typical-ly for direct cortical stimulation and the cognitive effects ofanesthesia that preclude extensive language testing. Theintracarotid sodium amobarbital (Wada) test is still consid-ered to be the gold standard technique for preoperativelanguage lateralization, but it has many drawbacks. Wadatesting is a relatively invasive procedure when compared tofMRI and is associated with non-negligible risks of seriouscomplications such as stroke. In fact, a recent study of morethan 600 patients undergoing Wada testing reported anoverall complication rate of 10.9 % [11]. In addition to thesafety concerns, Wada testing is also more uncomfortablefor the patient and is associated with a significantly highercost than functional MRI [12]. Many studies in the last twodecades have compared the relative language hemisphericlateralizing capability of Wada testing to that of fMRI. Withfew exceptions, these studies have consistently demonstrat-ed agreement between the two methods, with concordancerates ranging from 61 to 100 % depending on the individualseries [8, 13–29]. fMRI is a noninvasive, reproducible,whole-brain mapping technique that can provide in the samescan session information regarding both language localiza-tion and lateralization, whereas Wada testing can only pro-vide lateralization information. Furthermore, fMRI isgenerally more widely available than other noninvasivepreoperative brain mapping techniques (e.g., magnetoence-phalography/magnetic source imaging or transcranial mag-netic stimulation) and also does not suffer from some of thelimitations of these other techniques in terms of spatialresolution and evaluation of deep cortical and subcorticalstructures. Although fMRI has not supplanted the Wada testfor memory lateralization, it has replaced the Wada test forlanguage lateralization at many centers, particularly for cer-tain patient populations. For example, at our institution, forbrain tumor presurgical evaluation, fMRI is often performedin lieu of Wada testing because eloquent cortical localizationis critical as hemispheric language lateralization in thispatient population. However, for temporal lobe epilepsypresurgical evaluation, where memory evaluation is oftenas important as or even more important than language eval-uation, Wada testing remains the preoperative test of choice.However, despite the widespread acceptance of fMRI inbrain tumor presurgical evaluation, the optimization of clin-ical fMRI for language lateralization remains an area ofactive investigation. In performing language mapping withBOLD fMRI, language lateralization typically is expressedby the lateralization index (LI). This quantity provides

information about the relative contributions to overall lan-guage activation provided by the two cerebral hemispheres.Usually, this quantity is expressed in terms of the number ofsuprathreshold active voxels in each hemisphere; thus, theLI results are generally largely dependent on the statisticalthreshold used in generating the activation maps.

The choice of the paradigm performed by a patient forlanguage mapping has also been demonstrated to affect thehemispheric dominance determination; in particular, verbalfluency tasks have been reported to provide a more lateral-ized activation than naming tasks [21, 30].

Finally, there can be substantial differences in lateralitymeasures obtained at the global hemisphere level and inspecific regions of interest due to spatial variability in lan-guage representation across the brain [31]. This heterogene-ity could be more pronounced in patient populations inwhich brain damage has occurred, particularly in patientswith tumors, because of impairment of the BOLD responsein tumoral and peritumoral areas due to neurovascularuncoupling and susceptibility artifacts caused by hemor-rhage and micromineralization [32].

A study previously published by our group has partiallyaddressed these issues when BOLD fMRI was used forpresurgical mapping in brain tumor patients, comparing LIamong six commonly used clinical fMRI language para-digms, calculated using both a threshold-dependent and athreshold-independent approach [33]. However, in thatstudy, LI was determined through the use of ROIs thatencompassed the entire cerebral hemispheres and notlanguage-specific functional and anatomical areas ofinterest.

We designed this study in order to compare the effective-ness of lateralization of different expressive, receptive, andsemantic clinically used fMRI paradigms in anatomicallyand functionally defined language-specific ROIs.

Methods

Participants

Forty-one patients with brain tumors were included in thestudy. Table 1 reports the demographic information withtumor location and histology for each patient, as well ashandedness scores, tumor dimensions as reported in officialradiology reports in our electronic medical record, the actualfMRI language paradigms used for each patient, and theclosest eloquent cortical region (i.e., functional area or ana-tomic gyrus displaying activation) to the tumor margins. Allpatients were hand-handed as determined by the EdinburghHandedness Inventory [34]. The handedness score is calcu-lated as LI0[(R−L)/(L+R)]×100, where L and R representthe number of left and right responses, respectively, in the

1016 Neuroradiology (2012) 54:1015–1025

Table 1 Clinical/demographic data for the patients included in the study

Age Sex Handednessscore (%)

Tumor location Tumor size (cm) Histology/tumor grade Paradigms Closest eloquent language cortexlocation with respect to the tumormargins

25 F +40 Left frontal lobe 5.5 AP×3.3 TR×4.2CC

Oligoastrocytoma grade II LC, SC,SWG, R

Left DLPFC lateral and inferior

61 F +60 Right parietal lobe 1.5 AP×1.6 TR×1.2CC

Oligodendroglioma grade II R, SWG SMA two gyri anterior

66 F +90 Right parietal lobe 2.9 AP×4.6 TR×3.6CC

Oligodendroglioma grade II R, SWG, SC Right DLPFC inferior and anterior

43 M +70 Left frontal lobe 0.7 AP×0.5 TR×1.0CC

Glioblastoma grade IV R, SWG, SC Left MFG lateral to the lesion


Oligodendroglioma grade II R, SWG, SC Left MFG anterior and posterior

42 M +100 Left temporallobe

7.2 AP×4.9 TR×4.9CC

Astrocytoma grade II R, SWG, LC Left WA area superior and posterior

29 M +70 Right parietallobe

3.3 AP×2.8 TR×2.7CC

Anaplastic astrocytomagrade III

R, SWG, SC No critical activation in the vicinity

27 M +100 Right cingulategyrus

5.8 AP×2.6 TR×2.8CC

Oligodendroglioma grade II R, SWG,SC, LC

SMA anterior and superior

34 F +70 Right frontal lobe 3.0 AP×3.7 TR×2.4CC


R, SC, LC No critical activation in the vicinity


Glioblastoma grade IV R, SC, LC,SWG

Left BA anterior, Left WA inferiorand posterior

69 F +70 Right parietallobe

2.3 AP×2.O TR×2.2CC

Oligodendroglioma grade III R, SC, LC,SWG

Expressive speech PMC anterior


Oligoastrocytoma grade II R, SC, LC,SWG

Left BA anterolateral and superior

71 M +60 Left occipitallobe

1.8 AP×2.0 TR×2.0CC

Glioblastoma grade IV R, SC, LC Left WA and MTG anterior andsuperior


Oligodendroglioma grade II R, SC, LC,SWG

Left DLPFC posterior


4.0 AP×2.0 TR×2.3CC

Astrocytoma grade II R, SC, LC Left ITG posterior


Glioblastoma grade IV R, SC, SWG Left BA anterior, Left DLPFC superior

71 M +60 Right temporallobe

3.1 AP×2.8 TR×1.0CC

Astrocytoma grade II R, SC Right MTG and STG (WA) lateral


8.0 AP×6.2 TR×6.4CC

Astrocytoma grade II R, SC, LC,SWG

Left WA posterior


Oligodendroglioma grade II R, SC, LC,SWG

Left BA and Left DLPFC anterior

57 M +80 Right temporallobe

5.0 AP×5.0 TR×4.2CC


No critical activation in the vicinity(possible NVU)



Left MFG anterior and superolateral


Astrocytoma grade II R, SC, LC,SWG

Left IFG lateral, left DLPFCsuperolateral

35 M +100 Right frontal lobe 2.4 AP×3.0 TR×2.8CC

Oligodendroglioma grade II R, SC, SWG No critical activation in the vicinity

18 M +80 Left frontal lobe 0.11 AP×0.12 TR×0.13 CC

Oligodendroglioma grade II R, SC, SWG Left IFG superior and anterior

39 M +50 Right frontalparietal lobe

1.5 AP×1.9 TR×1.9CC

Astrocytoma grade II R, SC, SWG No critical activation in the vicinity

55 M +100 Left temporaloccipital lobe

4.2 AP×3.3 TR×2.9CC


SC, LC,SWG

Left MTG superior and anterior


6.3 AP×5.2 TR×3.0CC


Left STG (WA) posterior and superior

68 F +80 Right frontal lobe 2.5 AP×2. 5 TR×2.6CC

Glioblastoma grade IV R, SWG Right IFG anterior (BA homologue)


6.8 AP×4.1 TR×3.9CC

Glioblastoma grade IV SC, LC,SWG

Left inferior temporal lobeposteroinferior

57 F +60 Left temporallobe

0.4 AP×0.2 TR×0.5CC


R, LC, SWG Left STG (WA) anterior and superior,MTG posterior and superior

55 F +90 Glioblastoma grade IV R, LC, SWG Left WA posterior

Neuroradiology (2012) 54:1015–1025 1017

handedness questionnaire. The histopathology reported low-grade (WHO grades I and II) and high-grade (WHO gradeIII or IV) tumors for 22 and 19 patients, respectively.Lesions were located in the left hemisphere in 29 patientsand in the right hemisphere in 12 patients. Twenty of thetumors were located in the frontal lobe and 21 in the tem-poral and parietal lobes. The study was approved by theInstitutional Review Board, which did not require patientinformed consent because the data were retrospectively andanonymously analyzed.

Imaging

Patients were imaged on a 3.0-T Siemens Trio MRI system(Siemens Medical Solutions, Erlangen, Germany) using a headmatrix coil. For BOLD fMRI, axial slices were acquired using aT2* EPI sequence (33 slices, 4 mm thick with 1-mm gapbetween slices). In-plane spatial resolution was 3.75×3.75 mm2, TR02,000 ms, TE030 ms, 90° flip angle, 24-cmfield of view, and 64×64 acquisition matrix. A volumetric T1-weightedMPRAGE acquisitionwas acquired after the injectionof gadolinium contrast agent and used as a high-resolutionanatomic reference frame for overlay of functional activationmaps. Parameters for T1-weighted 3D MPRAGE sequenceused in the study are the following: TR02,300 ms, TI0900 ms, TE03.5 ms, 9° flip angle, 24-cm field of view, 256×

256 matrix acquisition, and slice thickness of 1 mm. Headmotion was minimized by the use of foam padding and straps.

fMRI paradigms

A Prism Acquire (Prism Clinical Imaging, Elm Grove, WI,USA) system was used for fMRI paradigm presentation. Fourblock design paradigms—rhyming (R), silent word generation(SWG), sentence completion (SC), and sentence listening com-prehension (LC)—were included for the purposes of this studyfrom a battery of clinically used language mapping tasks. Thefirst three of these were presented via a multimedia projector.The LC task was presented via headphones. For LC and R,30 seconds (s) of control taskwas followed by 30 s of active taskwith three cycles, for a total paradigm duration of 3 minutes(min) for each task. Twenty seconds of control task followed by20 s of active task with four cycles, for a total paradigm durationof 4 min, were used for SWG and SC. All paradigms werecovertly performed to minimize head motion. SWG was per-formed by 37 patients, R by 39, SC by 26, and LC by 30.

In SWG, patients were presented with a letter for 10 s andasked to generate words beginning with that letter during theactive block. The control block consisted of fixation on non-sense drawings. In R, rhyming and non-rhyming word pairswere presented every 3 s during the active block, whereaspaired bar patterns that matched or did not exactly match were

Table 1 (continued)

Age Sex Handednessscore (%)

Tumor location Tumor size (cm) Histology/tumor grade Paradigms Closest eloquent language cortexlocation with respect to the tumormargins

Left temporallobe

2.3 AP×2.3 TR×1.2CC

72 M +30 Left parietal lobe 2.2 AP×2.7 TR×1.3CC

Glioblastoma grade IV R, LC, SWG Parietal receptive inferior


2.3 AP×3.3 TR×1.9CC

Ganglioglioma grade II R, LC, SWG Left STG superior and anterior

44 M +40 Right frontal lobe 5. 2 AP×5.0 TR×4.5CC

Oligodendroglioma grade II R, LC, SWG No critical activation in the vicinity


Oligodendroglioma grade II R, LC, SWG Left BA anterior, WA superoposterior


3.4 AP×2.9 TR×1.6CC

Fibrillary astrocytoma gradeII

R, LC, SWG Left STG (WA) posterior and superior


Anaplasticoligodendroglioma gradeIII

R, LC, SWG Left DLPFC superior


3.4 AP×3.5 TR×2.2CC

Infilitrating astrocytomagrade II

R, LC, SWG Left STG superior and posterior


3.7 AP×3.1 TR×1.7CC

Oligodendroglioma grade II R, LC, SWG Left STG (WA) superior


Glioblastoma grade IV R, LC, SWG Left IFG (BA) anterior

36 F +70 Left frontal lobe 2.40 AP×3.1 TR×3.1 CC

Glioblastoma grade IV R, LC, SWG Left SMA medial, DLPFC lateraland inferior

M male, F female, IFG inferior frontal gyrus, MFG middle frontal gyrus, ITG inferior temporal gyrus, MTG middle temporal gyrus, STG superiortemporal gyrus, BA Broca’s area, WAWernicke’s area, SMA supplementary motor area, DLPFC dorsolateral prefrontal cortex, NVU neurovascularuncoupling, AP anterior posterior, TR transverse, CC craniocaudal

1018 Neuroradiology (2012) 54:1015–1025

presented every 3 s during the control block. Patients wereasked to determine whether words rhymed or not in the activecondition and to determine whether the bar patterns matchedor did not match in the control condition. Task performancewas assessed by asking the patients to indicate a rhymingword pair or matching bar pattern pair by a button response.In SC, patients were shown strings of nonsense symbolsresembling letters during the control block and asked simplyto guide their eyes over them. During the active block, realsentences with the last word missing were presented every 5 sand patients were instructed to think of a word to completeeach presented sentence. In LC, the active condition consistedof sentences aurally presented every 5 s, and patients wereasked to determine whether or not each sentence represented atrue statement; the patients were instructed to press a buttonfor only affirmative responses. The control condition involvedlistening to garbled (backward) speech without button presses.For each patient, the imaging session was preceded by atraining session outside the MRI scanner in order to familiar-ize the patient with the tasks, assess the patient’s ability toadequately perform the tasks, and confirm that the patientunderstood the instructions for all tasks.

Image analysis

Images were processed and analyzed using Analysis ofFunctional NeuroImages software [35]. Raw data were first

assembled into a 4D dataset such that the BOLD signal timeseries was reconstructed for each voxel. Each time serieswas then temporally interpolated so that all the slices in thesame volume were acquired at the same time. Subsequently,all the volumes were spatially aligned using a rigid bodyalgorithm and then co-registered with the structural T1images using an affine registration algorithm [36]. The twotransformations were combined to minimize the amount ofinterpolation applied to the data. For each task, activationmaps were generated using the general linear model analy-sis. Each voxel was assigned a T score revealing the corre-lation between an expected hemodynamic response functionand the BOLD signal time series. The statistical thresholdapplied varied for each paradigm for each patient in order tobalance between type I (false positive) and type II errors(false negative), and this threshold was decided by consen-sus by two raters. ROI analysis was performed for LI cal-culation. Two functional and two anatomical ROIs weremanually drawn by each rater (JN, GD). The functionalROIs (Fig. 1) were defined for each paradigm by the acti-vation cluster which most closely approximated the positionof BA and WA. Two anatomical ROIs (Fig. 2) were thendefined by the boundaries segmenting the frontal lobe asone volume (“expressive area” or EA) and segmenting thecombination of the temporal and parietal lobes (“receptivearea” or RA) as the second volume using the central sulcusas the anatomic landmark defining both the posterior margin

Fig. 1 Example of functional language ROIs. a Language activationmap for a silent SWG task thresholded at 3.0 (t statistics, with 3 voxelcluster spatial extent threshold) overlaid on T1 post-contrast 3DMPRAGE images for a 69-year-old female patient included in thestudy with grade III oligodendroglioma in the right parietal lobe. Note

the strong left hemispheric dominance and the activation in both frontaland temporal–parietal lobes. b Functional BA in the dominant hemi-sphere ROI (cyan) with its contralateral mirror ROI (green) and func-tional WA in the dominant hemisphere ROI (yellow) with itscontralateral mirror ROI (red) as drawn by one rater

Neuroradiology (2012) 54:1015–1025 1019

of the EA ROI and anterior margin of the RA ROI. This wasdone in an effort to encompass expressive and receptiveactivation clusters in the EA and RA ROIs, respectively,based on the lobar anatomy rather than simply the locationsof individual activation cluster centers. Contralateral mirrorROIs were also defined both for the functional and anatom-ical ROIs. Please note that functionally defined ROIs in theleft hemisphere were large enough to not exclude criticalactivation in the right hemisphere in their mirrored contra-lateral ROIs. A threshold-independent LI was determinedby comparing the integrated T score (value) weighted(weight0T score2) distributions of all positively task-correlated voxels between the left and the right hemi-spheric homologous ROIs [37]. In particular, the areasunder the weighted distributions were computed and LIwas defined as

LI ¼ LH � RH

LH þ RH

where LH denotes the integrated weighted distribution for theleft ROI and RH for the right hemisphere ROI. This formulayields values between −1 and +1, which are positive for lefthemispheric dominance and negative for right dominance.Please note that although the actual LI computation methodis threshold-independent because it takes into account thedistribution of the t values of all voxels included in the ROIs,the actual choice of statistical threshold for functional ROIdelineation was determined by consensus between the two

raters. Only the mean values for LI computed based on inde-pendent ROI tracings by the two raters were used for furtherstatistical analysis. Please also note that we defined bilateralityas an absolute value of LI<0.1, as established in multiplepreviously published fMRI studies of language lateralization[33, 38].

Statistical methods

An inter-rater agreement was evaluated on a sample of fourdatasets (which was also used as a training set to familiarizethe two raters with the ROI tracing methodology) to assessthe reproducibility of LI measurement among the two raters(JN, GD). A further more extensive evaluation of inter-rateragreement was performed on all of the datasets for allparadigms and ROIs obtained in this study. An intra-classcorrelation coefficient (ICC) was computed for both sam-ples. An ICC value of 0.83 (implying very good inter-raterreliability) was obtained for the training set, and an overallICC for the entire study of 0.66 was obtained (suggestinggood overall inter-rater reliability). An ANOVA test withspecific post hoc test was performed using SPSS v. 17.0software in each ROI for four different groups (the para-digms) to determine significant differences among the para-digms in the determination of language lateralizationincluding multiple factors (tumor grade, tumor lobar loca-tion, and hemisphere). Statistical significance was consid-ered at the p<0.05 level.

Fig. 2 Example of anatomical language ROIs. a Anatomical left (red)hemisphere expressive ROI (EA) with its mirror right hemisphere(yellow) ROI. b Left (red) hemisphere receptive ROI (RA) with itsmirror right hemisphere (yellow) ROI as drawn by one rater on the T1

post-contrast 3D MPRAGE images for the same patient in Fig. 1. Thecentral sulcus was used as the anatomic landmark to separate EA andRA ROIs

1020 Neuroradiology (2012) 54:1015–1025

Results

In Table 2, the main findings of statistical ANOVA arereported. In Fig. 3, the distributions of the LI for the fourdifferent paradigms in the four different ROIs are repre-sented as the mean±standard deviation. A main effect wasfound for the paradigms (p<0.04) in the BA ROI. Post hoctests revealed a significantly higher mean LI value for SWG(0.36±0.25) compared to LC (0.16±0.25, p00.009) and forR (0.40±0.22) compared to LC (0.16±0.25, p00.001). Nomain effect was found for any of the other factors includedin the ANOVA or significant interaction among them. Inthe WA ROI, the analysis did not reveal any main orinteraction effect.

A main paradigm effect was present in the EA ROI (p<0.04), with the SWG mean LI (0.28±0.19) significantlyhigher than the LC LI (0.12±0.16, p00.01). In this ROI,the analysis also reported a main effect on tumor grade;specifically, the mean LI in low-grade tumors (0.27±0.19)was significantly higher than the mean LI in high-gradetumors (0.15±0.16, p00.001). Comparison of the two pop-ulations at a single paradigm level demonstrated higher LIfor SWG in low-grade lesions (0.35±0.16) compared tohigh-grade lesions (0.20±0.18, p00.01) as well as higherLI for R in low-grade lesions (0.26±0.13) compared to high-grade lesions (0.15±0.15, p00.02). No interaction effectswere present among the ANOVA factors except for a trendin the lobe×grade interaction (p00.055).

A main effect for the paradigms was present in the RAROI (p00.03), but post hoc tests did not reveal any signif-icant pair effect difference. No other main or interactioneffects were found for the remaining factors.

Discussion

The identification of the dominant hemisphere for languagerepresentation is essential in the preoperative risk assessmentof patients with lesions located in the vicinity of expectedeloquent language cortex. When such eloquent cortex isresected from the dominant hemisphere for language repre-sentation, the chances of a long-term postoperative language

deficit are higher than if the resection involved the non-dominant hemisphere. Hemispheric dominance also affectsthe intraoperative cortical stimulation mapping strategy and,ultimately, the overall therapeutic decision making (i.e., grosstotal resection vs. debulking vs. open biopsy with conserva-tive medical management).

Multiple studies have compared the language lateraliza-tion results provided by fMRI word generation tasks withthe Wada test results [15, 16, 19, 20, 22, 23]. Overall, agenerally excellent agreement, between 83.3 and 100 %, hasbeen reported, and these findings did not vary significantlydepending on whether the lateralization analysis was fo-cused on the entire brain or on specific brain regions (front-al, parietal, temporal). A rhyming task was compared to theWada test in a study by Baciu et al. [24]; they foundlateralization concordance in 94 % of the cases using aglobal approach for LI computation. The comparison be-tween the Wada test results and fMRI using semantic lan-guage tasks has also been thoroughly investigated, bothusing holohemispheric and ROI approaches for the assess-ment of laterality. For example, in studies using a holohemi-spheric approach in epilepsy patients, Binder et al. [13]found 100 % concordance in 22 cases for a semantic deci-sion task and Carpentier et al. [25] reported concordancebetween the fMRI LI scores and the Wada test in 8 of 10patients using both auditory and visual stimulus presentationduring language tasks, with the visual tasks providinghigher lateralization. Similarly, Szaflarski et al. [26] reporteda high correlation (r<0.75, p<0.001) between the fMRI andthe Wada LI scores for a semantic decision task using aglobal language region of interest including functional Bro-ca’s and Wernicke’s areas. A variable agreement (from 63 to100 %) was instead found in a study by Baciu et al. [27]using four different semantic tasks (category judgment,word stem completion, and semantic and phonological as-sociation). Spreer et al. [28] demonstrated that regional LIcorresponded better than global LI with the Wada test, andconcordance between fMRI LI and the Wada test wasreported in all cases in the frontal areas and in all but twocases in the temporal areas. Similarly, a better agreementbetween frontal LI than temporal LI was found in a study byBenke et al. [29].

Table 2 Statistical distribution of LI for four paradigms (SWG, R, SC, and LC) in four different ROIs (BA, WA, EA, and RA)

BA WA EA RA

Paradigm Mean SD Paradigm Mean SD Paradigm Mean SD Paradigm Mean SD

SWG 0.36 0.25 SWG 0.45 0.21 SWG 0.28 0.19 SWG 0.23 0.12

R 0.40 0.22 R 0.38 0.26 R 0.21 0.15 R 0.17 0.12

SC 0.23 0.30 SC 0.46 0.26 SC 0.23 0.21 SC 0.22 0.17

LC 0.16 0.24 LC 0.44 0.21 LC 0.12 0.16 LC 0.19 0.11

Neuroradiology (2012) 54:1015–1025 1021

All the cited studies demonstrate that the paradigms chosenin this study have been extensively validated for languagelateralization by a direct comparison with the results of the“gold standard” Wada test. Therefore, fMRI provides an ex-cellent noninvasive alternative to the more invasive Wada testand can potentially replace it for preoperative assessment oflanguage lateralization. However, more research is needed todetermine which fMRI paradigms are most reliable for spe-cific patient populations. To our knowledge, this is the firststudy assessing how different language paradigms and differ-ent ROIs affect the determination of language hemisphericdominance in brain tumor patients. In a group of right-handedbrain tumor patients, we investigated the relative effectivenessof four different commonly clinically used fMRI paradigms inlateralizing language function in various brain regions con-taining language-eloquent cortex. We used a threshold-independent approach for LI computation in order to avoidany dependency of the LI on arbitrarily chosen statisticalthresholds following ROI selection. Furthermore, the lateral-ization results obtained with this relatively novel method of LIcomputation have already been validated via comparison tothe Wada testing results in a group of epilepsy patients [38].

The results of this study demonstrate that SWG and R arethe best among this group of paradigms for the lateralizationof expressive language areas, in agreement with our previ-ous study where holohemispheric ROIs, excluding the cer-ebellum and the occipital lobe, were used for LIcomputation [33]. Although higher mean LI values werereported in that paper based on threshold-dependent analy-sis, for the threshold-independent analysis (comparable tothe technique used in the current study), lower mean valueswere reported for SWG and R (21.68±14.74 and 20.35±11.13, respectively) than those reported in our current studyfor the BA, EA, and also, surprisingly, WA ROIs [33]. It isencouraging that our functional and anatomic ROI-basedmethods yielded higher LIs using the SWG and R tasks thanour previously published holohemispheric approach to theevaluation of language lateralization. This improved lateral-ization likely reflects the overall greater functional specific-ity of the regions included in the ROI analyses. Thesefindings confirm the good lateralizing capability of verbalfluency and, more generally, expressive language tasks inthis particular cohort of patients. However, the receptive andsemantic paradigms used in this study were poor lateralizers

Fig. 3 Summary of the study statistics. Distribution of the LI for thefour different paradigms analyzed in this study—silent word genera-tion (SWG), rhyming (R), sentence completion (SC), and sentencelistening comprehension (LC)—in the four different ROIs: BA (a),EA (b), WA (c), and RA (d). SWG and R are better lateralizing tasks

than SC and LC in the expressive ROIs (BA and EA), whereas the tworeceptive tasks did not perform better than SWG and R in the receptiveROIs (WA and RA). The results are represented as the mean±standarddeviation

1022 Neuroradiology (2012) 54:1015–1025

both in the BA and EA ROIs. This is somewhat surprisingsince the LC task involved keypad button presses with thedominant right hand in the task blocks that were not coun-terbalanced with responses in the control blocks; this para-digm was deliberately designed in this fashion in order toelicit activation in the primary motor cortex (PMC), andespecially the face representation area of the PMC that iscritical for expressive language/speech, and thus would havebeen expected to result in higher EA LIs. Instead, these tasksdemonstrated good lateralization only in the WA ROIs, butnot in the RA ROIs. Furthermore, in both these regions, theydid not perform better than the R and SWG for the determi-nation of lateralization because there were no significantdifferences in the LI values. These results are consistentwith similar studies that compared word generation andsemantic tasks in epilepsy patients [16, 26].

It could be argued that it may not be useful to includesemantic or receptive tasks in performing presurgical map-ping by BOLD fMRI, at least in brain tumor patients, basedon these results. However, a relatively newly imaging-validated dual-stream theory of language representationdescribes a larger number of cortical regions and connectingwhite matter tracts beyond the classical description of BAand WA that are involved in language processing. Thistheory describes a unilateral and generally strongly left-lateralized pathway for the dorsal stream, which is involvedin expressive/productive language processing, and a some-what less left-lateralized and more bilateral pathway for theventral stream, which is more involved in receptive andsemantic language processing, thus explaining why recep-tive and semantic tasks may provide a less lateralized pat-tern of activation than expressive language tasks [39].

The ANOVA performed in this study demonstrated aneffect for tumor grade in the EA, whereas tumor hemispher-ic or lobar location did not demonstrate any similar effect onthe lateralization determination. In particular, our best ex-pressive lateralizing tasks, SWG and R, yielded significantlylower LI in patients with grade III and IV lesions comparedto grades I and II. In the group of right-handed patientsincluded in this study, about two in three of the patientshad lesions in the left language-dominant hemisphere. Thedecrease of BOLD contrast in the proximity of high-gradehyperperfused tumors due to the uncoupling between neu-ronal activation and vascular reactivity is a phenomenonwell documented in the literature and may represent themain reason for the decrease in LI in this subgroup ofpatients for generally strongly lateralizing tasks such asSWG and R [32, 40]. A recent paper by Jiang et al. [41]has described this phenomenon of abnormally decreasedvasoactivity in the vicinity of primary brain tumors (asassessed by BOLD response following carbogen inhalation)and its associated effect of asymmetrically reduced ipsile-sional sensorimotor activation.

LI was in general higher for all tasks in the functionallydefined ROIs (i.e., BA and WA ROIs). This is not surprisingconsidering that these ROIs were drawn starting from theclusters of activations in the left hemisphere, and then thecontralateral ROIs were mirrored. The wider spatial ROIsdefined on the gyral anatomy may include regions of non-essential (or participatory) activation, which may be lesslateralized for language function. However, SWG demon-strated good lateralizing capability in all ROIs, which is inagreement with the results of a meta-analysis recently per-formed on the studies comparing the agreement between theWada test results and fMRI that demonstrated the bestresults for word generation tasks [42].

However, as useful as hemispheric language lateraliza-tion with fMRI has been for clinical surgical decision mak-ing, it does have some limitations. While contralesionalhemispheric lateralization does give a neurosurgeon moreconfidence in proceeding with tumor resection, it must bekept in mind that localization (i.e., proximity of eloquentcortex to tumor margins) is just as important a considerationas lateralization. In addition, despite overall single hemi-spheric dominance, variable contributions of the contralat-eral non-dominant hemisphere to language function arepossible. Some of the cortical regions that display activationin the non-dominant (right) hemisphere may, indeed, behighly eloquent, and, as the observed variability of the LIvalues in this cohort (all LIs<1) suggests, varying degrees ofbilateral true language representation exist in the humanbrain.

The main limitation of this study is the lack of compar-ison of the fMRI lateralization results with those of the goldstandard Wada test. However, our cohort of patients con-sisted exclusively of right-handed individuals, and in arecent study, it has been demonstrated that 95 % of dextralshave left language hemispheric dominance. Furthermore, inour institution, the Wada test is not routinely performed inthese brain tumor patients because of its invasiveness andthe need for eloquent cortical localization as well as lan-guage lateralization (which fMRI can provide but Wadatesting cannot); therefore, such data would not be available.The manual segmentation of both functional and anatomicalROIs could certainly introduce an operator dependencyissue in this method to some extent. An atlas-based ap-proach would eliminate this operator dependency by auto-matic selection of the ROIs, but brain tumor patients oftenpresent with distorted anatomy, making the warping of thebrain into a stereotactic space technically difficult toachieve. However, we demonstrated that with a clear defi-nition of the criteria to define the ROIs, a very good inter-rater agreement was obtained. The application of circum-scribed ROIs around Broca’s and Wernicke’s areas in ourprotocol could represent another limitation of this study. Ithas been shown indeed that language areas can be found

Neuroradiology (2012) 54:1015–1025 1023

quite remotely from their classical localizations, especiallyin pathological states such as epilepsy [43]; therefore, wemay have incurred the risk of missing intrahemisphericallyshifted, i.e., reorganized, language activation. However, todate, little is known about the phenomenon of postsurgi-cal language plasticity in brain tumor patients, and forthis reason, more studies are needed to investigate thistopic [44].

Conclusion

Our study confirms that the silent word generation task and,more generally, the expressive paradigms in general allowthe determination of the overall hemispheric language later-alization, as well as lateralization within key regions knownfor their essential role in language processing, in braintumor patients. In such a patient population, we have dem-onstrated the lateralizing capability of these paradigms de-spite the presence of several confounding variables, such asneurovascular uncoupling and cortical reorganization thatare known to affect BOLD fMRI activation maps in thesetting of brain neoplasia. It remains difficult to assesslanguage lateralization in receptive language areas withfunctional imaging, both because language comprehensionappears to represent an inherently less well-lateralized func-tion than speech production in the human brain and becausemore effective receptive and semantic paradigms need to bedeveloped in the future.

This study may represent an important contribution to-ward the standardization of paradigms for presurgical lan-guage mapping, which will be necessary for the promotionof the more widespread clinical utilization of BOLD fMRIbeyond established academic centers.

Acknowledgements This work is partially and indirectly funded bySiemens Medical Solutions (grant no. 39154). Siemens played no rolein the study design, data collection and analysis, or preparation of themanuscript.

Conflict of interest We declare that we have no conflict of interest.

References

1. Roux FE, Boulanouar K, Ranjeva JP, Manelfe C, Tremoulet M etal (1999) Cortical intraoperative stimulation in brain tumors as atool to evaluate spatial data from motor functional MRI. InvestRadiol 34:225–229

2. Ruge MI, Victor J, Hosain S, Correa DD, Relkin NR et al (1999)Concordance between functional magnetic resonance imaging andintraoperative language mapping. Stereotact Funct Neurosurg72:95–102

3. Hirsch J, Ruge MI, Kim KH, Correa DD, Victor JD et al (2000) Anintegrated functional magnetic resonance imaging procedure for

preoperative mapping of cortical areas associated with tactile,motor, language, and visual functions. Neurosurgery 47:711–721

4. Brannen JH, Badie B, Moritz CH, Quigley M et al (2001) Reli-ability of functional MR imaging with word-generation tasks formapping Broca’s area. AJNR Am J Neuroradiol 22:1711–1718

5. Sunaert S, Yousry TA (2001) Clinical applications of functionalmagnetic resonance imaging. Neuroimaging Clin N Am 11(2):221–236

6. Yetkin FZ, Mueller WM, Morris GL, McAuliffe TL, Ulmer JL,Cox RW, Daniels DL, Haughton VM (1997) Functional MR acti-vation correlated with intraoperative cortical mapping. AJNR Am JNeuroradiol 18(7):1311–1315

7. Moritz C, Haughton V (2003) Functional MR imaging: paradigmsfor clinical preoperative mapping. Magn Reson Imaging Clin NAm 11(4):529–542

8. Rutten GJ, Ramsey NF, van Rijen PC, Noordmans HJ, van VeelenCW (2002) Development of a functional magnetic resonance im-aging protocol for intraoperative localization of critical temporo-parietal language areas. Ann Neurol 51(3):350–360

9. Giussani C, Roux FE, Ojemann J, Sganzerla EP, Pirillo D, Papagno C(2010) Is preoperative functional magnetic resonance imaging reli-able for language areas mapping in brain tumor surgery? Review oflanguage functional magnetic resonance imaging and direct corticalstimulation correlation studies. Neurosurgery 66(1):113–120

10. Bizzi A, Blasi V, Falini A, Ferroli P, Cadioli M, Danesi U, AquinoD, Marras C, Caldiroli D, Broggi G (2008) Presurgical functionalMR imaging of language and motor functions: validation withintraoperative electrocortical mapping. Radiology 248(2):579–589

11. Loddenkemper T, Morris HH, Moddel G (2008) Complicationsduring the Wada test. Epilepsy Behav 13:551–553

12. Medina LS, Aguirre E, Bernal B, Altman NR (2004) FunctionalMR imaging versus Wada test for evaluation of language lateral-ization: cost analysis. Radiology 230:49–54

13. Binder JR, Swanson SJ, Hammeke TA, Morris GL, Mueller WMet al (1996) Determination of language dominance using functionalMRI: a comparison with the Wada test. Neurology 46:978–984

14. Hertz-Pannier L, Gaillard WD, Mott SH, Cuenod CA, BookheimerSY et al (1997) Noninvasive assessment of language dominance inchildren and adolescents with functional MRI: a preliminary study.Neurology 48:1003–1012

15. Gaillard WD, Balsamo L, Xu B, Grandin CB, Braniecki SH et al(2002) Language dominance in partial epilepsy patients identifiedwith and fMRI reading task. Neurology 59:256–265

16. Lehéricy S, Cohen L, Bazin B, Samson S, Giacomini E et al (2000)Functional MR evaluation of temporal and frontal language dom-inance compared to the Wada test. Neurology 54:1625–1633

17. Desmond JE, Sum JM, Wagner AD, Demb JB, Shear PK et al(1995) Functional MRI measurement of language lateralization inWada-tested patients. Brain 118:1411–1419

18. Benbadis SR, Binder JR, Swanson SJ, Fischer M, Hammeke TA etal (1998) Is speech arrest during Wada testing a valid method fordetermining hemispheric representation of language? Brain Lang65:441–446

19. Bahn MM, Lin W, Silbergeld DL, Miller JW, Kuppusamy K et al(1997) Localization of language cortices by functional MR imag-ing compared with intracarotid amobarbital hemispheric sedation.AJR Am J Roentgenol 169:575–579

20. Yetkin FZ, Swanson S, Fischer M, Akansel G, Morris G et al(1998) Functional MR of frontal lobe activation: comparison withWada language results. AJNR Am J Neuroradiol 19:1095–1098

21. Benson RR, FitzGerald DB, LeSueur LL, Kennedy DN, Kwong KKet al (1999) Language dominance determined by whole brain func-tional MRI in patients with brain lesions. Neurology 52:798–809

22. Woermann FG, Jokeit H, Luerding R, Freitag H, Schulz R et al(2003) Language lateralization by Wada test and fMRI in 100patients with epilepsy. Neurology 61:699–701

1024 Neuroradiology (2012) 54:1015–1025

23. Sabbah P, Chassoux F, Leveque C, Landre E, Baudoin-Chial S et al(2003) Functional MR imaging in assessment of language domi-nance in epileptic patients. NeuroImage 18:460–467

24. Baciu M, Kahane P, Minotti L, Charnallet A, David D et al (2001)Functional MRI assessment of the hemispheric predominance forlanguage in epileptic patients using a simple rhyme detection task.Epileptic Disord 3:117–124

25. Carpentier A, Pugh KR, Westerveld M, Studholme C, Skrinjar O etal (2001) Functional MRI of language processing: dependence oninput modality and temporal lobe epilepsy. Epilepsia 42:1241–1254

26. Szaflarski JP, Holland SK, Jacola LM, Lindsell C, Privitera MD etal (2008) Comprehensive presurgical functional MRI languageevaluation in adult patients with epilepsy. Epilepsy Behav 12:74–83

27. Baciu MV, Watson JM, Maccotta L, McDermott KB, Buckner RLet al (2005) Evaluating functional MRI procedures for assessinghemispheric language dominance in neurosurgical patients. Neu-roradiology 47:835–844

28. Spreer J, Arnold S, Quiske A, Wohlfarth R, Ziyeh S et al (2002)Determination of hemisphere dominance for language: comparisonof frontal and temporal fMRI activation with intracarotid amytaltesting. Neuroradiology 44:467–474

29. Benke T, Köylü B, Visani P, Karner E, Brenneis C et al (2006)Language lateralization in temporal lobe epilepsy: a comparisonbetween fMRI and the Wada test. Epilepsia 47:1308–1319

30. Petrovich NM, Holodny AI, Brennan CW, Gutin PH (2004) Iso-lated translocation of Wernicke’s area to the right hemisphere in a62-year-man with a temporo-parietal glioma. AJNR Am J Neuro-radiol 25:130–133

31. Seghier ML, Kherif F, Josse G, Price CJ (2011) Regional andhemispheric determinants of language laterality: implications forpreoperative fMRI. Hum Brain Mapp 32:1602–1614

32. Holodny AI, Schulder M, Liu WC, Wolko J, Maldjian JA, KalninAJ (2000) The effect of brain tumors on BOLD functional MRimaging activation in the adjacent motor cortex: implications forimage-guided neurosurgery. AJNR Am J Neuroradiol 21:1415–1422

33. Pillai JJ, Zaca D (2011) Relative utility for hemispheric lateraliza-tion of different clinical fMRI activation tasks within a compre-hensive language paradigm battery in brain tumor patients as

assessed by both threshold-dependent and threshold-independentanalysis methods. NeuroImage 54:S136–S145

34. Oldfield RC (1971) The assessment and analysis of handedness:the Edinburgh Inventory. Neuropsychologia 9:97–113

35. Cox RW (1996) AFNI: software for analysis and visualization offunctional magnetic resonance neuroimages. Comput Biomed ResInt J 29:162–173

36. Saad ZS, Glen DR, Chen G, Beauchamp MS, Desai R, Cox RW(2009) A new method for improving functional-to-structural MRIalignment using local Pearson correlation. NeuroImage 44:839–848

37. Branco DM, Suarez RO, Whalen S, O’Shea JP, Nelson AP et al(2006) Functional MRI of memory in the hippocampus: lateralityindicies may be more meaningful if calculated from whole voxeldistributions. NeuroImage 32:592–602

38. Suarez RO, Whalen S, Nelson AP, Tie Y, Meadows ME et al(2009) Threshold-independent functional MRI determination oflanguage dominance: a validation study against clinical gold stand-ards. Epilepsy Behav 16:288–297

39. Saur D, Kreher BW, Schnell S, Kümmerer D, Kellmeyer P et al(2008) Ventral and dorsal pathways for language. Proc Natl AcadSci USA 105:18035–18040

40. Hou BJ, Bradbury M, Peck KK, Petrovich NM, Gutin PH et al(2006) Effect of brain tumor neovasculature defined by rCBV onBOLD fMRI activation volume in the primary motor cortex. Neu-roImage 32:489–497

41. Jiang Z, Krainik A, David O, Salon C, Troprès I, Hoffmann D,Pannetier N, Barbier EL, Bombìn ER, Warnking J, Pasteris C,Chabardes S, Berger F, Grand S, Segebarth C, Gay E, Le Bas JF(2010) Impaired fMRI activation in patients with primary braintumors. NeuroImage 52(2):538–548

42. Dym RJ, Burns J, Freeman K, Lipton ML (2011) Is functional MRimaging assessment of hemispheric language dominance as goodas the Wada test. Radiology 216:446–455

43. Ojemann G, Ojemann J, Lettich E, Berger M (1989) Corticallanguage localization in left, dominant hemisphere. An electricalstimulation mapping investigation in 117 patients. J Neurosurg71:316–326, Republished: J Neurosurg. 108:411–421

44. Pillai JJ (2010) Insights into adult postlesional language corticalplasticity provided by cerebral blood oxygen level-dependentfunctional MR imaging. AJNR Am J Neuroradiol 31:990–996

Neuroradiology (2012) 54:1015–1025 1025

Documents

Effectiveness of four different clinical fMRI paradigms for preoperative regional determination of language lateralization in patients with brain tumors