Functional MRI in Epilepsy

Functional MRI in Epilepsy

*Jeffrey R. Binder, †Eric Achten, ‡R. Todd Constable, §John A. Detre,\William D. Gaillard,¶Clifford R. Jack, and **David W. Loring

*Department of Neurology, Medical College of Wisconsin, Milwaukee, Wisconsin, U.S.A.; †University Hospital Gent, Gent,Belgium; ‡Departments of Diagnostic Radiology and Neurosurgery, Yale University, New Haven, Connecticut; §Department ofNeurology, University of Pennsylvania, Philadelphia, Pennsylvania;\Clinical Epilepsy Section, National Institutes of Health,

Bethesda, Maryland; ¶Department of Radiology, Mayo Clinic, Rochester, Minnesota; and **Department of Neurology, MedicalCollege of Georgia, Augusta, Georgia, U.S.A.

The discovery of blood oxygenation level–dependent(BOLD) contrast in magnetic resonance imaging (MRI)created exciting prospects for advancing our knowledgeabout the neural organization of cognitive operations inthe brain and for improving on available techniques forfunctional mapping in neurologic patients (1). We pre-sent here a brief review of functional MRI (fMRI) meth-ods and applications, with emphasis on identification ofmotor, language, and memory systems that are at risk inpatients treated surgically for intractable epilepsy.

FMRI DATA AND IMAGE ANALYSIS

The most commonly used contrast mechanism em-ployed in fMRI is BOLD contrast (1). BOLD contrast isdependent on changes in blood flow and so is not a directmeasure of neural activity. The physiologic basis ofBOLD contrast was actually identified in the 1980s bypositron emission tomography (PET) researchers (2). Af-ter initiation of a functional activation task, those areas ofthe brain that are actively engaged by the task experiencea local increase in cerebral blood flow. The magnitude ofthe blood flow increase exceeds the associated increasein local cerebral oxygen consumption. This leads to aspatially localized increase in the ratio of oxyhemoglobinto deoxyhemoglobin when compared with that in thenonactive state. Oxyhemoglobin is diamagnetic (i.e., itseffect on T2-relaxation rate is fairly negligible), whereasdeoxyhemoglobin is paramagnetic (i.e., it accelerates therate of T2 relaxation) (3). This, therefore, provides a linkbetween physiologic events that occur during mental ac-tivation and localized changes in signal intensity that aredetectable by MRI. The magnitude of the activation-dependent change in MR signal, however, is quite mod-est. Activation tasks that produce a quite vigorousresponse, such as sensorimotor stimulation or visual

stimulation via flashing lights, produce only a 1–5%change in signal at 1.5 Tesla. Regional changes in bloodflow require several seconds to develop, making the tem-poral resolution of fMRI relatively coarse compared withunderlying neural activity.

The method by which a functional activation map isderived from an fMRI examination is driven in large partby the quite modest activation signal observed with theBOLD contrast mechanism. During fMRI, the subjectperforms an activation task while a series of MR imagessensitized to BOLD contrast is acquired over time (Fig.1). The task in question could be active (i.e., the subjectcould be asked to memorize something, squeeze asponge), or it could be passive (i.e., the palm of thesubject could be brushed). An fMRI examination is typi-cally divided into periods in which images are acquiredwhile the activation task is being applied and periods inwhich images are acquired during a foil task, or nonac-tivation period. Localized signal-intensity changes in thebrain that occur in response to the activation are ex-tracted from this time series of images in one of severalways. The most widely used approach is cross-correlation (4). The temporal relation between the activeand the foil epochs of the activation paradigm and thetiming of each individual MR image are known a priori.A cross-correlation analysis is performed to identifythose pixels whose signal intensity fluctuates in phasewith the known timing of the activation task. A timinglag between onset of the task and the cross-correlationfunction is incorporated to accommodate the hemody-namic nature of BOLD contrast (5). Pixels whose corre-lation with the task exceeds a certain threshold areincluded in the final activation map, whereas pixelswhose temporal fluctuation in signal intensity is not cor-related with onset and offset of the activation task areeliminated. The functional activation map is then over-laid on matching slices from anatomic images of thebrain (Fig. 2).

Alternative methods have been used to extract “acti-vation” from an fMRI time series. Instead of cycling an

Address correspondence and reprint requests to Dr. J. R. Binder atDepartment of Neurology, Medical College of Wisconsin, 9200 W.Wisconsin Avenue, Milwaukee, WI, 53226-3522, U.S.A. E-mail:[email protected]

Epilepsia,43(Suppl 1):51–63, 2002Blackwell Publishing© International League Against Epilepsy

51

activation task on and off repeatedly, the paradigm mayconsist of simply a single activation period and a singlenonactivation period. A statistical test then asks the ques-tion, which pixels had a signal intensity that differedsignificantly during the acquisition of the “on” MR im-ages versus those MR images that were acquired duringthe “off” period. Another increasingly popular alterna-tive is event-related fMRI (6). With this method, a singlebrief stimulus event is administered either at periodic orirregular intervals during a free-running acquisition offMRI images. This differs from the cross-correlationmethod in that with a correlation method, the active andrest portions of the fMRI paradigm are typically from 10to 40 s in duration and of equal length.

All methods of fMRI data processing ultimately sharethe common feature that a statistical criterion must beselected to threshold the activation map. That is, certainpixels in the image are deemed “significantly” activatedand others are deemed not to have been activated by thetask. Unfortunately, there is no way to determine whatstatistical thresholding criterion should be used to obtaina map of “true” activation. Simply varying the magnitudeof the correlation coefficient will result in apparentlystrikingly different activation maps (Fig. 3). Activationmaps produced by fMRI data analysis are all, to someextent, dependent on somewhat arbitrary selection of sta-tistical thresholding criteria.

A major problem with functional MRI is subject mo-tion (7) (Fig. 4). Consider a voxel at the edge of thebrain. If the patient’s head moves just one-tenth of apixel (0.3–0.4 mm), roughly a 10% change in the signalintensity can result. This dwarfs the magnitude of thedetectable BOLD response for most activation para-digms. If the patient happens to move in coincidencewith onset and offset of the activation paradigm, this willresult in movement-related signal-intensity changes thatfar exceed the typical BOLD response but appear to becorrelated with the task paradigm. Yet 0.3–0.4 mm of

head motion is quite small, and, for many patients, mo-tions of this magnitude are difficult to avoid. A great dealof attention, therefore, has been paid to methods of mo-tion correction. The most common approach is retrospec-tive realignment of MR images by voxel-based matchingprograms (8,9). The difficulty with this approach is thatthe spin-excitation history may be altered, and the tem-poral features of the activation paradigm may not beencoded correctly in the fMRI time series. This informa-tion cannot be recovered if it has never been encoded inthe original fMRI data. An alternative to this approach isreal-time detection and prospective correction of bulkhead motion during the fMRI acquisition (10–12).

FIG. 1. Functional magnetic resonance imaging (fMRI) acquisi-tion. The temporal relations among the fMRI activation task (i.e.,the stimulus), the resulting physiologic blood oxygenation level–dependent (BOLD) response, and the acquisition of echo-planarimaging images in a typical fMRI time series is illustrated.

FIG. 3. Correlation coefficient threshold effects. The three im-ages are functional magnetic resonance imaging (fMRI) activa-tion maps from the same time-series data of a bilateral hand-tapping activation task. Note that simply changing the correlationcoefficient threshold value has a marked effect on the appear-ance of the activation pattern.

FIG. 2. How functional magnetic resonance imaging (fMRI) im-ages are formed. Top left: An unthresholded gray-scale image ofcorrelation coefficients from a bilateral hand-tapping fMRI activa-tion experiment. Top right: An AFNI printout of the signal-intensity time course of a voxel from the motor strip. Note how thetime-course data on the bottom of this figure correlate with thesinusoidal reference function of the induced activation task. Bot-tom left: The result of thresholding the activation map at r = 0.5.Bottom right: A composite image formed by superimposing thethresholded activation map onto the matching anatomic MRIslice.

J. R. BINDER ET AL.52

Epilepsia, Vol. 43, Suppl 1, 2002

IMAGE ACQUISITION

There are many issues to consider when designing anfMRI experiment, and the proper choice of image-acquisition parameters requires recognition of thesources of noise and artifacts and the methods availablefor reducing such artifacts. Much attention has been di-rected toward determining the optimal echo time (TE),flip angle, and pulse sequence to use in fMRI studies ata range of field strengths, and choosing the optimal val-ues for these parameters is fairly straightforward. Todate, little attention has been directed to determining theoptimal repetition time (TR), but recent evidence indi-cates that minimizing the TR maximizes statistical power(13).

Two persistent problems with fMRI, image distortionand signal loss, arise from the presence of static fieldinhomogeneities, and these problems worsen with in-creasing field strength. When adjacent tissues have dif-ferent magnetic susceptibilities, a nonuniform magneticfield is produced. These field nonuniformities cause geo-metric image distortion and signal loss. The field inho-mogeneities have high spatial variability and tend to belocalized in different regions of the brain such that theycannot be eliminated through main-field shimming. Thelocations of these inhomogeneities have been well docu-mented (14), and in general, they are most pronounced inthe basal and medial temporal regions of the brain.

The first step to take in minimizing field inhomoge-

neities involves shimming over the brain volume to pro-vide the most uniform magnetic field possible.Generally, however, global shimming is not sufficient toremove the local signal variations and will therefore noteliminate the image-distortion or signal-loss problems.Localized shimming on a specific region of interesteliminates the problem in that region but typically de-grades the shim in other regions, thereby abrogating thepossibility of imaging more than a single region of in-terest. In the following paragraphs, we separately addressapproaches to solving the image-distortion and signal-loss problems.

Image distortion can be reduced during acquisition bychanging the manner in which the data are acquired, buttrade-offs are associated with each approach. Imaging athigher bandwidth reduces distortion but decreases thesignal-to-noise ratio of the individual images. Imagingwith multishot sequences reduces the image distortion indirect proportion to the number of shots, but the effectiveTR increases linearly, thereby reducing statistical powerand temporal resolution. Reducing the acquisition matrixsize also reduces image distortion but at the expense ofspatial resolution. Postprocessing methods also can beapplied to reduce further or even to eliminate image dis-tortion. These include field mapping (15,16) or point-spread function mapping (17) to measure the distortionand then correct for it, and registration algorithms (18)that find the best fit of the distorted functional images toa nondistorted anatomic image. Eliminating these geo-metric image distortions is crucial for precise mapping ofthe functional data onto high-resolution anatomic scans.

The problem of signal loss can be severe and prohibiteffective imaging in basal temporal regions. This prob-lem is particularly difficult to eliminate because theBOLD contrast (1,19) we wish to observe is based ondifferences in local magnetic susceptibility that occur asfresh oxygenated blood replaces deoxygenated blood onbrain activation, leading to a small localized increase inMR signal intensity. If all sensitivity to susceptibilityeffects is removed, sensitivity to the BOLD effect alsowill be eliminated. Strategies are required, then, thatminimize the macroscopic static field susceptibility ef-fects while maintaining sensitivity to microscopic fieldeffects. Approaches to this problem include tailored rf-pulses (20–22), z-shimming (23–28), and thin-slice ac-quisitions (29–31). The easiest of these to implement isthe thin-slice approach, although this is dependent on rfpulses that can produce excellent slice profiles and onlong TRs to allow many slice locations to be imaged.Z-shimming is highly effective but requires an increasein the effective TR by an amount dependent on the num-ber of shims selected. Tailored rf-pulses could poten-tially refocus all through-plane dephasing because offield inhomogeneities in a single acquisition, but evencurrent state-of-the-art imaging hardware is insufficient

FIG. 4. Effect of motion. These images were obtained during apresurgical functional magnetic resonance imaging (fMRI) map-ping study in a patient with a right perirolandic low-grade tumor(arrows). Top row: The images were the result of an fMRI run inwhich the patient moved her head. Bottom row: Bottom row: Theimages are at roughly the same anatomic position from an fMRIrun without discernible head motion in the same fMRI examina-tion. Note that clear-cut activation demarcating the central sulcusis not discernible in the motion-degraded activation maps in thetop row. In the motion-free fMRI run illustrated in the bottom row,the position of the central sulcus relative to the tumor is clearlyshown.

FUNCTIONAL MRI 53


to play out the complex pulses needed to make this ap-proach practical. It is clear that there are many trade-offsto be considered in determining the appropriate acquisi-tion parameters and that selection of these parametersdepends on the specific purpose to which fMRI is beingapplied.

MAPPING SENSORIMOTOR SYSTEMS

Patients with lesions around the central sulcus oftenhave epilepsy. The questions we ask when confrontedwith such a patient with intractable epilepsy and a lesionaround the rolandic sulcus are as follows: can the lesionor abnormal brain tissue be removed/treated without in-creasing the neurologic deficit; is the intervention goingto improve the patient’s epilepsy; and can these ques-tions be answered with fMRI? During the surgical plan-ning for such patients with lesions in the frontal andparietal lobe, it is therefore important to map the prox-imity or involvement of the primary motor and sensorycortex (Figs. 5 and 6). These critical areas border thecentral sulcus (CS), and detection of this anatomic struc-ture is important as well (32). The CS can be identifiedon anatomic imaging studies if several anatomic land-marks are recognized (33–35), although a rather low in-terobserver variability has been reported (36). Pathologicconditions like tumors might alter the location of thecentral sulcus and the related eloquent areas, and in casesof dysplasia, trauma, or ischemia, the critical zones forcertain functions may be shifted to other brain areas (37).Surgical exploration of such lesions to alleviate associ-

ated intractable epilepsy requires knowledge of the loca-tion of bordering brain functions to provide a betteroutcome, preferably without increasing the neurologicdeficit (32,38–40).

Many studies have addressed the presurgical use offMRI to identify the CS and other critical cortex in pa-tients with tumors or arteriovenous malformations (41–54). Some have investigated tumor proximity to the CSand compared the fMRI localization of the sensorimotorcortex with invasive cortical stimulation (41,42,44,45,49,52–54). A general conclusion was that noninvasivepreoperative mapping of sensorimotor cortex allowed abetter prepared surgical approach and better patient in-formation (Fig. 6).

Performing fMRI of the motor and sensory system isstraightforward most of the time and can be performedon most commercial MRI scanners. Simple movement ofbody parts will robustly activate somatosensory and mo-tor cortex: MI, SI, and SII With more complex para-digms, the supplementary motor area and cerebellarareas also are activated (55). Somatotopic activation hasbeen demonstrated by many groups (56–58).

Problems that can be expected during acquisition offMRI data include stimulus-correlated head motion (7),inability to move because of existing neurologic deficit,and possible altered hemodynamic response function dueto disturbed blood-flow regulation (41). Stimulus-correlated head motion should be addressed with the pa-tient. A practice session must be organized, during whichthe task movements are exercised. Such a training ses-sion, together with mild restraint of the head, is usuallysufficient to reduce motion artifacts to an acceptablelevel. When movement is impaired by a lesion, compli-ance with simple movement paradigms requires a greater

FIG. 5. Patient with simple partial motor seizures and paresthe-sias in the left leg. Magnetic resonance imaging (MRI) showed anarteriovenous malformation in the right rolandic region (white ar-rows). On both figures, fMRI activation by simultaneous move-ment of both hands is depicted as red spots. Note that therepresentation of the contralateral hand is not at the expectedposition. A: Ipsilateral sensorimotor foot activation (moving theright foot) in green and sensory activation (leg rubbing) in blue. B:The same activations for the contralateral leg in the same colors.Sensorimotor activation of the contralateral leg on B is displacedlaterally and is fainter than the activation of the ipsilateral leg onA. Clinical examination showed no performance difference formovement of the legs. Pure sensory stimulation of the contraleralleg did not yield clear activation in the SI region, while bilateral SIIactivation is visible for both legs. Movement of the hand and thefeet also activates SII.

FIG. 6. Patient with complex partial epilepsy and a right rolandiccavernous hemangioma (arrowheads). The patient did not haveany functional deficit. Superimposed in color on the anatomicimages are the functional magnetic resonance imaging (fMRI)movement-activation zones (thresholded at p<0.05, corrected) ofboth hands (axial slices) and of left hand alone (coronal slices).The activated zones of the left hand are very close to the lesion.This information proved to be very valuable for the neurosurgeon,and the approach to the lesion was altered to avoid damage tothe sensorimotor cortex of the hand.



effort by the patient, which usually results in a strongerfunctional activation in the supplementary motor area orthe posterior parietal lobule (54). When the neurologicdeficit makes movement of a body part impossible, pas-sive movement can be used to provide proprioceptivestimulation. Although MI will not activate, premotor andsensory areas will if these functions are not impaired aswell. A change in the hemodynamic response functioncan be expected with arteriovenous malformations andvascular tumors with strong contrast enhancement (Fig.5). Lack of normal activation should raise the suspicionof altered blood-flow regulation, making normal BOLDresponses impossible.

The image acquisition can be done in single or mul-tislice gradient-echo imaging (54) or with echo-planarimaging. Because more than one area of critical cortexcan border the lesion, combined paradigms using event-related and block designs can be used in a single fMRIsession to assess language functions, auditory cortex, andthe somatosensory and motor system (59). The examplesin Figs. 5 and 6 show patient data from the UniversityHospital Gent, Belgium, performed at 1 T. Simple move-ment of limbs was used as the activation paradigm versusrest. The functional maps prove that even at 1 T, excel-lent results can be obtained. Data analysis for these fig-ures was performed with the open source software“SPM99” (http://www.fil.ion.ucl.ac.uk/spm/), but sev-eral other packages also are available (60).

MAPPING LANGUAGE SYSTEMS

The aim of localizing language functions is to mini-mize postoperative language deficits, such as anomia,that can result from epilepsy surgery (61). fMRI has beenused extensively to study normal language processing,although the areas identified in different studies havevaried markedly, likely owing to use of differentlanguage-activation tasks, control tasks, imaging tech-niques, and data-processing methods (62). Language isnot a single process, but rather involves specialized sen-sory systems for speech, text, and object recognition;access to whole-word information; access to word mean-ing; processing of syntax; and multiple mechanisms forwritten and spoken language production. Neuropsycho-

logical studies suggest a certain modularity of organiza-tion of these language subsystems (63), and it is unlikelythat any single activation procedure could identify all ofthem. Some commonly used task combinations and thebrain regions they typically “activate” are given in Ta-ble 1.

A few examples may be illustrative of some mainissues in task design. Numerous studies over the pastdecade have shown that hearing words—whether thetask involves passive listening, repeating, or categoriz-ing—activates the superior temporal gyrus bilaterallywhen compared with a resting state (Fig. 7A) (64–67).The symmetry of this activation may be surprising, but aconsideration of the task contrast (complex sounds com-pared with no sounds) reveals that the “rest” baselinecontains no controls for primary or secondary auditoryprocesses that engage auditory cortex in both superiortemporal gyri (68). Thus, much of the activation pro-duced by contrasting word-listening with no sound couldbe due to prelinguistic auditory processing. These acti-vation patterns bear almost no relation to language domi-nance measured by Wada testing (69). Another problemin using such a paradigm for language mapping is thatleft-lateralized brain areas associated with semantic pro-cessing are probably activated during the “rest” state(70). Subtraction of this activity reduces sensitivity fordetecting activation in these regions by the word-listening task.

Similar problems occur in designs that contrast read-ing or naming tasks with a resting or visual-fixationbaseline. The baseline condition contains little in the wayof controls for prelinguistic visual form-recognition pro-cesses in ventral occipitotemporal regions, which thusdominate the activation map. Benson et al. (71) foundthat such procedures do not reliably produce lateralizedactivation and do not correlate with language dominancemeasured by Wada testing.

More widely used than listening, repeating, reading,and object-naming tasks are “word generation” tasks(also called fluency tasks) that require word retrieval inresponse to a verbal cue. Subjects are given a beginningletter, a semantic category, or a word, and must retrievea phonologically or semantically associated word. Thistask strongly activates the dominant inferior and dorso-

TABLE 1. Some task contrasts used for language mapping and the regions in which robust activations are typically observed

Dorsolateralprefrontal

Superiorprefrontal

Superiortemporal

Ventrolateraltemporal

Ventraloccipital

Angulargyrus

Hearing Sentences vs. Rest BReading Sentences vs. Rest B L > R BObject Naming vs. Rest B L > R BSemantic Decision vs. Nonsemantic Task L L L LWord Generation vs. Rest L > R L > R BWord Generation vs. Reading L

L, left hemisphere; R, right hemisphere; B, bilateral.

FUNCTIONAL MRI 55


lateral frontal lobe, including prefrontal and premotorareas (64,72–74), and lateralization measures obtainedfrom this frontal activation agree well with Wada lan-guage lateralization (69,71,75–77). Posterior languageareas such as middle and inferior temporal gyri, fusiformgyrus, and angular gyrus are only weakly activated bythe word-generation task compared with a resting state ora word-reading control (64,72–74). Word-generationtasks are usually performed silently in fMRI studies toavoid movement artifacts. The resulting absence of task-performance data is usually not a problem when clearactivation is observed, but bars the investigator from as-sessing the contribution of poor task performance incases with poor activation.

Another approach involves pairing a word-compre-hension task with a nonlinguistic control task. In theprotocol used by Binder et al. (78,79), subjects catego-rize auditory words based on semantic criteria (“Is itfound in the United States?”) (78,79). This language taskis compared with a tone-discrimination task that controlsfor primary auditory, attentional, working-memory, andmotor aspects of the language task. The resulting activa-tion pattern is strongly left-lateralized and involves bothprefrontal and posterior association areas (Fig. 7B).Demb et al. (80) developed a similar procedure in whichsemantic categorization of words is contrasted with anonsemantic uppercase/lowercase discrimination. Al-though posterior regions have not yet been studied withthis paradigm, prefrontal activation is strongly left-lateralized. Both of these task combinations produce lat-eralized activation in individual epilepsy patients that isstrongly correlated with Wada language lateralization(78,81). An attractive feature of these paradigms is thatmeasured behavioral responses, consisting of simple but-ton presses for stimuli that meet response criteria, permittask performance to be precisely quantified.

Although several task paradigms produce measures oflanguage lateralization that correlate with Wada lan-guage scores, there is reason to be cautious in replacingthe Wada language test with fMRI. First, only a fewatypical or crossed-dominant individuals have been stud-ied with both Wada and fMRI. This is the very conditionthat is perhaps the most important to detect, and the smallnumbers currently available do not allow any firm con-clusions to be drawn about sensitivity or specificity ofthe various fMRI tests. Second, the incidence of signifi-cant discrepancy between fMRI and Wada lateralizationmeasures is not known, nor have the reasons for theoccasional discrepancies been investigated. Third, it ispossible—even likely—that the degree of lateralizationdepends on the particular task contrast used, and it is notyet clear which task or set of tasks is optimal.

In addition to lateralization information, fMRI pro-vides relatively detailed maps of intrahemispheric acti-vation foci. These maps could one day assist surgeons inplanning the boundaries of a resection, but such an ap-plication seems scientifically insupportable at present,for several reasons. Regions not activated by one lan-guage task may be activated by another. Thus, lack ofactivation in a given region does not necessarily indicatelack of an important functional role for the region. Con-versely, some regions activated during language tasksmay play a minor, supportive role rather than a criticalrole in language, and resection of these active foci maynot necessarily produce clinically relevant deficits. Sev-eral reports suggest a greater than chance correspon-dence between fMRI activation foci and language fociidentified during awake cortical stimulation mapping(82,83). How these two techniques compare in terms ofaccuracy of predicting critical language sites, and wheth-er fMRI alone is sufficient for identifying these sites, areproblems for future research.

MAPPING EPISODIC MEMORY SYSTEMS

The primary goal of fMRI mapping of memory func-tion in temporal lobe epilepsy (TLE) is to determine theextent to which the affected versus unaffected temporallobes are engaged during memory function. This infor-mation is presumably of use in determining the risks ofamnestic complications from temporal lobectomy. If pro-spective studies clearly demonstrate that resection in aregion of fMRI activation results in a decrement inmemory performance, fMRI data might further be usedin planning specific resections in individual cases toavoid postsurgical amnesia. Because memory function issubserved by the same brain regions that typically harborthe seizure focus itself, memory-activation patterns ob-served with fMRI also may contribute to seizure lateral-ization and to prediction of seizure-free outcome fromtemporal lobectomy, as has been demonstrated for the

FIG. 7. Group average functional magnetic resonance imaging(fMRI) activation patterns in neurologically normal, right-handedvolunteers during two language paradigms. A: Listening to spo-ken words contrasted with resting (28 subjects). Superior tempo-ral activation occurs bilaterally. B: Semantic decision on auditorywords contrasted with a tone decision control task (30 subjects).Activation is strongly left-lateralized in multiple prefrontal andsensory association cortices. The images are serial axial sectionsspaced at 15-mm intervals through stereotaxic space, starting atz = –15. The left hemisphere is on the reader’s left. Green linesindicate stereotaxic x and y axes.



memory portion of the Wada test (84,85) and interictalmetabolic imaging (86,87).

The selective intracarotid amobarbital, or Wada, testwas initially developed to determine lateralization ofspeech dominance (88), and was subsequently extendedto evaluate lateralized asymmetries in memory functionin an effort to predict additionally the occurrence of am-nestic syndromes after resections (89). Although theWada test has been the gold standard for preoperativelateralization of language and memory function, it is in-vasive and provides only a limited period for testing.fMRI potentially provides a noninvasive alternative ap-proach to presurgical memory lateralization and localiza-tion that can be repeated if the findings are inconclusive.

As a first step in validating fMRI memory lateraliza-tion as a clinically useful tool, it is logical to attempt tocompare fMRI memory lateralization with that obtainedby Wada testing. Preliminary results suggest thatmemory lateralization findings by these two modalitiesagree to a large extent (90). However, because fMRIexamines endogenous function whereas the Wada test isfundamentally a lesion study, it also is reasonable toexpect that these modalities may differ, and that the find-ings obtained with each modality may be complementaryrather than entirely duplicative (91). Reductions in fMRImemory activation also appear to lateralize seizure focicorrectly in the majority of cases (92), and very prelimi-nary results also suggest that postsurgical amnesia cor-relates with fMRI activation ipsilateral to the resection(93).

Episodic memory is typically engaged during fMRI byexplicit or incidental encoding tasks, with subsequentrecognition testing to verify encoding success. Whereasthe hippocampus proper is the brain region most com-monly associated with episodic memory function (94),fMRI studies have often demonstrated activation moreposteriorly in the parahippocampal formation. The ex-

planation for this is poorly understood, but growing evi-dence suggests that it may be a sensitivity issue becausespecial shimming procedures (27) and signal averagingacross multiple subjects more reliably yield activation inthe hippocampus proper (J.A. Detre, personal communi-cation; see Fig. 8).

Encoding of complex visual scenes produces bilateralmedial temporal lobe activation (95) and results in asym-metric activation in patients with unilateral TLE (90)(see Fig. 8). More closely to approximate the Wada test(96), material-specific stimuli also may be used, al-though if normal activation is unilateral, some internalreference will be required to calibrate the observed acti-vation in patients. The use of a blocked design maxi-mizes sensitivity and minimizes the need for accuratemodels of when encoding actually takes place duringstimulus processing and of the hemodynamic response.In contrast, event-related designs allow segregation ofsuccessfully and unsuccessfully encoded trials (97),which may be useful in matching patients and controlsfor task performance.

Complex cognitive functions such as memory typi-cally activate a distributed network of brain regions, notall of which are critical for task performance. Con-versely, some critical regions may not be activated be-cause of limitations in sensitivity or task design. fMRIstudies of memory in TLE represent an ideal clinicalapplication for this new modality and provide an excel-lent model for validating fMRI more generally. TLE is arelatively stereotyped disorder that is confined to a smallbrain region; therefore only a limited number of activa-tion paradigms must be developed to assess its function.Furthermore, because temporal lobectomy is clinicallyindicated in many patients with TLE regardless of fMRIfindings, it provides an opportunity to determine the con-sequences of resection in or near a region of focal acti-vation. As with any new diagnostic or therapeutic

FIG. 8. Functional magnetic resonance imaging (fMRI) of episodic memory in temporal lobe epilepsy (TLE). Left: Examples of complexvisual scenes and control images used during a scene-encoding task. The control is a single randomly retiled image. Subjects are askedto remember the scenes for subsequent testing. Middle: Group-activation map (n = 17) using a random-effects model showing activationin visual association cortex and extending into the medial temporal lobe and hippocampus (arrows). Right: Suprathreshold activation ina medial temporal region of interest in patients with right (top ) and left (bottom) TLE. Decreased activation is observed in the regionipsilateral to the seizure focus (arrows).

FUNCTIONAL MRI 57


modality, fMRI memory localization should be carefullyvalidated in a prospective fashion before using it in rou-tine clinical management.

fMRI MAPPING IN CHILDREN

Focal epilepsy amenable to surgery is a pediatric dis-ease with a mean onset at 7 years (98). Neuronal injurybefore age 3 years and early age at seizure onset areassociated with transfer of language function to the typi-cally nondominant hemisphere. Later injury is associatedwith varying degrees of reorganization to homotopiccontralateral or adjacent regions in ipsilateral hemisphere(99–102). fMRI has been used in a number of studies toidentify language dominance in adults (69,71,77,78,102).Comparison studies with electrocortical stimulationmapping (ESM) suggest a very good correspondence be-tween activation with blood flow and disruption by cor-tical stimulation for motor and language tasks (82,103,104). As with adults, fMRI can be used to evaluatechildren with localized epilepsy (105,106).

Children as young as 5 years old may be studied suc-cessfully with cognitive tasks (107,108). As with adults,motion and cooperation are critical. Tasks must beadapted for cognitive ability. Choice of experimentalconditions is particularly important, as control conditionscommonly used for adults (for reading: pseudowords,letter strings, or false fonts; for auditory comprehension:speech played backward or presented in an unfamiliarlanguage) may be interpreted as language by youngerchildren, thus obscuring activation signal (109). Issuesgermane to challenges in pediatric fMRI have recentlybeen reviewed (108,110).

The BOLD response appears to be similar in adultsand children. Investigators, however, often find a need touse varying statistical thresholds in children and in pe-diatric patients (105,111–113). It is unclear whether thisreflects a technical consideration in imaging children ora physiologic phenomenon. There is conflicting evidencethat (a) the activation signal is more diffuse and lesspronounced in children, and (b) that the activation signalis stronger, with greater blood flow per gray-matter voxelreflecting greater synaptic density in children (108,112).For example, there is some evidence that activation dur-ing verbal-fluency tasks in children is somewhat morewidespread and may be less strongly lateralized than thatin adults because of greater homotopic activation in thenondominant hemisphere (112), but other studies havefound less activation with similar degrees of laterality(109,114). Studies that compare activation maps inadults and children find activation patterns are funda-mentally similar in location and laterality (108,112,115).Language networks for verbal fluency, auditory compre-hension, and reading comprehension appear to bestrongly lateralized and focal after age 7 years in normal

volunteers (109,112,114,116,117). Preliminary evidencesuggests that in some children, activation maps arehighly lateralized and focal by age 5 years (108).

Verbal fluency to categories or letters (and also verbgeneration to nouns) show similar activation patterns(118,119), which is important as young children performthe former task more readily. Verbal fluency and seman-tic decision tasks typically produce 1.5 to 9 times greateractivation (i.e., number of activated voxels) in the dom-inant prefrontal cortex than in the nondominant hemi-sphere. They are the most commonly used tasks forevaluation of pediatric epilepsy surgery candidates, and,as in adults, show general agreement with Wada testingand ESM (105–107,111,119–121). Tasks involving read-ing or auditory comprehension are better suited for iden-tifying language in dominant temporal lobe (Fig. 9).Tasks of reading stories or naming to description (“whatis a yellow fruit?” answer: “banana”) show highly later-alized activity in dominant mid- and superior temporalgyrus as well as inferior frontal gyrus (IFG) and mesialfrontal gyrus (MFG) (109,116) and have been used inpatients as young as 8 years (122). Auditory-based com-prehension paradigms also have been used, either listen-ing to stories or auditory naming to description; they alsoreliably identify temporal language cortex in agreementwith Wada testing (117,123). Auditory paradigms alsomay be used for preliterate young or cognitively im-paired children. Object-naming tasks have not provedreliable in children (111,119).

There are∼60 fMRI studies reported in children withepilepsy ranging from age 8 to 17 years (105,111,119,122–124). When compared with Wada testing, as inadults, there is general agreement; however, there is in-complete agreement in∼5% of cases, in which one mo-dality suggests bilateral language representation, and theother, unilateral language representation. It is uncertainwhich provides the correct information, as Wada testingand fMRI usually do not test the same aspects of lan-guage. fMRI can be used in the large majority of patientsto identify hemispheric dominance and can be used as aguide for ESM. Atypical or negative studies must beinterpreted with caution. fMRI has the advantage of be-ing more friendly to some children than the Wada test,can be used to assess a variety of language functions, andcan be repeated easily (113,121). Cognitively impairedchildren may not tolerate fMRI, but they may also havegreater difficulty with Wada testing and ESM. UnlikePET, fMRI does not confer radiation, and normative dataare available for comparison. A practical limitation is thelower age limit at which fMRI studies may be performed.Seven years is the lower limit for most cognitive studiesin children, but studies can be performed in well-motivated or properly trained children as young as 5 andperhaps 4 years. fMRI has the ability not only to later-alize and localize language functions in children, but also



to trace the impact of chronic disease, such as epilepsy,on the organization of cognitive networks during devel-opment.

fMRI COMPARED WITH WADA TESTING

One primary clinical goal of fMRI is to displace Wadatesting as the standard of care for determining languageand memory dominance in candidates for epilepsy sur-gery. If established as a valid and reliable technique,fMRI will either render the Wada test obsolete, or at least

reduce its role to a secondary procedure to be used onlywhen fMRI is not practical because of either technicalconsiderations or patient variables.

A significant advantage of fMRI is that it is noninva-sive. The Wada test, in contrast, typically involves thetransfemoral catheterization of the carotid arteries fol-lowed by the introduction of a short-acting barbiturate,typically sodium amobarbital. After injection, the ante-rior two thirds of the hemisphere is anesthetized for sev-eral minutes while language and memory are tested.Acute drug effects may produce behavioral confounds ofsedation and agitation, which can compromise validity ofindividual Wada results. fMRI avoids these confoundingeffects as well as the effects of aphasia on memory as-sessment. At least theoretically, fMRI also can separatethe different memory stages of registration, encoding,and retrieval. Because fMRI is noninvasive, it avoids therisks associated with invasive arteriography and, whennecessary, can be repeated easily.

For fMRI to displace the Wada as the gold standard, itwill be necessary to establish its equivalence to the Wadatest, which has been repeatedly validated with respect tolanguage representation, memory function, and predic-tion of both cognitive and seizure outcome. What appearto be small differences in approach, however, may besufficiently large to prevent fMRI from completely re-placing the Wada test as the first-line approach for lan-guage and memory assessment. Even with fMRIlanguage testing, which is more fully developed thanfMRI memory protocols, there are reports of outcomediscrepancies. For example, Wada testing in a patientwith left posterior temporal–inferior parietal tumor indi-cated left cerebral language dominance, which was con-firmed with cortical stimulation mapping. Using alanguage fMRI paradigm that had been validated inhealthy controls, however, right cerebral language domi-nance was inferred (125). Thus, there is the potentialwith our current understanding of fMRI and language forerrors that could result in significant language morbidityif relying on fMRI language studies alone.

fMRI involves activation procedures that, in theory,will identify multiple regions in a network involved withtask participation. Participation by a region, however,does not imply that the region is critical for task perfor-mance. In addition, regions that are not “activated” onfMRI may be critical. For example, motor tasks do nottypically activate subcortical motor regions, although ob-viously a lesion of the corticospinal tract will lead tomotor weakness. The Wada test assesses each hemi-sphere in isolation and uses a pharmacologic lesion thatis designed to model the surgical lesion effects. By doingso, the Wada test evaluates the independent contributionof the temporal lobe to be included during anterior tem-poral lobectomy by examining the memory results afterthe injection contralateral to the seizure focus (functional

FIG. 9. Functional magnetic resonance imaging (fMRI) study inan 8-year-old, right-handed boy with a left temporal lobe seizurefocus. Left image is left brain. Four tasks are performed, eachwith a boxcar design of six cycles; each cycle is 64 s, dividedbetween task and a control condition (for reading, the control isviewing dots; for auditory paradigms, the control is silent rest). Allparadigms are unmonitored and covert. Top row: Read re-sponse naming (RRN); task items are reading a brief descriptionof an object that the child then silently names (e.g., “What is along yellow fruit?” answer: “banana.”) New cues are providedevery 4 s. Second row: Reading a fable (Table); during this task,the child reads a story adjusted for reading level. Third row:Auditory response naming (ARN); the auditory version of the firsttask, in which the child listens to a three- to five-word descriptionof an object. Fourth row: Listening to stories (Listen); for thistask, children listen to stories compared with rest. All studiesshow strong left lateralization of language processing. Writtenand auditory processing occur in similar regions in dominant(here left) temporal lobe. Reading tasks evoke greater dominant(again left) frontal activation, especially when word retrieval isrequired. These patterns are typical for left hemisphere–dominant children and adults.

FUNCTIONAL MRI 59


hippocampal adequacy) and the contribution of thesingle temporal region that will be relied on for memoryafter surgery by examining memory results after injec-tion ipsilateral to seizure onset (functional hippocampalreserve) (126). Further, interhemispheric differencescores provide better prognostic information regardingmemory outcome than relying simply on ipsilateral orcontralateral Wada scores alone (127).

Many investigators have applied different statisticalthresholds to determine which regions are “significant”in different subjects undergoing fMRI. In such cases, weare left with the difficulty of validating a procedure inwhich different thresholds are selected, based on indi-vidual patient characteristics. Because statistical signifi-cance does not imply clinical significance, and clinicalsignificance is not always reflected in statistical results, itis difficult to determine what the relation is between“activation,” which is somewhat arbitrarily defined andvariable, and the clinical significance of those findings.

The problem of fMRI memory validation is mademore difficult when considering the different experimen-tal protocols used to perform fMRI activations. As wasthe case with the Wada during its refinement, it is likelythat one set of fMRI stimuli will not be as robust asanother set, and currently no single Wada procedure isagreed on as the most appropriate protocol for assessingmemory. Thus, one difficulty will be that fMRI, a gen-eral procedure rather than a standardized test, must bedemonstrated to be equivalent to the Wada, another gen-eral procedure rather than a standardized test. Variabilitywithin the two procedures will lengthen and increase thenumber of validation procedures needed to replace Wadatesting with fMRI scanning. Ultimately, however, thevalidation of fMRI will depend not directly on its rela-tion to the Wada test, but on its relation to the factors thatare predicted by Wada testing: postoperative outcomes.This combination of comparisons will be necessary be-fore fMRI results are treated with the same confidence asWada results.

REFERENCES

1. Ogawa S, Lee TM, Kay CR, et al. Brain magnetic resonanceimaging with contrast dependent on blood oxygenation.Proc NatlAcad Sci U S A1990;87:9868–72.

2. Fox PT, Raichle ME. Focal physiological uncoupling of cerebralblood flow and oxidative metabolism during somatosensorystimulation in human subjects.Proc Natl Acad Sci U S A1986;83:1140–4.

3. Turner R, LeBihan D, Moonen CTW, et al. Echo-planar timecourse MRI of cat brain oxygenation changes.Magn Reson Med1991;22:159–66.

4. Bandettini PA, Jesmanowicz A, Wong EC, Hyde JS. Processingstrategies for time-course data sets in functional MRI of the hu-man brain.Magn Reson Med1993;30:161–73.

5. Lee AT, Glover GH, Meyer CH. Discrimination of large venousvessels in time-course spiral blood-oxygen-level-dependent mag-netic resonance functional neuroimaging.Magn Reson Med1995;33:745–4.

6. Buckner RL, Bandettini PA, O’Craven KM, et al. Detection ofcortical activation during averaged single trials of a cognitive taskusing functional magnetic resonance imaging.Proc Natl Acad SciU S A1996;93:14878–83.

7. Hajnal JV, Myers R, Oatridge A, et al. Artifacts due to stimuluscorrelated motion in functional imaging of the brain.Magn ResonMed 1994;31:283–91.

8. Cox RW. AFNI: software for analysis and visualization of func-tional magnetic resonance neuroimages.Comput Biomed Res1996;29:162–73.

9. Friston KJ, Williams S, Howard R, et al. Movement-related ef-fects in fMRI time-series.Magn Reson Med1996;35:346–55.

10. Lee CC, Jack CR, Grimm RC, et al. Real-time adaptive motioncorrection in functional MRI.Magn Reson Med1996;36:436–44.

11. Lee CC, Grimm RC, Manduca A, et al. A prospective approachto correct for inter-image head rotation in fMRI.Magn ResonMed 1998;39:234–43.

12. Ward HA, Riederer SJ, Grimm RC, et al. Prospective multi-axialmotion correction for fMRI.Magn Reson Med2000;43:459–69.

13. Constable RT, Spencer DD. Repetition time in echo planar func-tional MR imaging.Magn Reson Med2001;46:748–55.

14. Ojemann JG, Akbudak E, Snyder AZ, et al. Anatomic localizationand quantitative analysis of gradient refocused echo-planar fMRIsusceptibility artifacts.Neuroimage1997;6:156–67.

15. Jezzard P, Balaban RS. Correction for geometric distortion inecho planar images from Bo-field variations.Magn Reson Med1995;34:65–73.

16. Reber PJ, Wong EC, Buxton RB, et al. Correction of off-resonance related distortion in EPI using EPI-based field maps.Magn Reson Med1998;39:328–30.

17. Robson M, Gore JC, Constable RT. Measurement of the pointspread function in MRI using constant time imaging.Magn ResonMed 1997;38:733–40.

18. Studholme C, Constable RT, Duncan JS. Non-rigid mapping offunctional EPI to anatomical MRI incorporating a spin echo im-age distortion model.IEEE Trans Med Imaging2001;19:1115–27.

19. Ogawa S, Lee TM, Nayak AS, et al. Oxygenation-sensitive con-trast in magnetic resonance image of rodent brain at high mag-netic fields. Magn Reson Med1990;14:68–78.

20. Cho ZH, Ro YM. Reduction of susceptibility artifact in gradient-echo imaging.Magn Reson Med1992;23:193–200.

21. Stenger VA, Boada FE, Noll DC. Three-dimensional tailored RFpulses for the reduction of susceptibility artifacts in T2*-weightedfunctional MRI.Magn Reson Med2000;44:525–31.

22. Chen N, Wyrwicz AM. Removal of intravoxel dephasing artifactin gradient echo images using a field-map based RF refocusingtechnique.Magn Reson Med1999;42:807–12.

23. Frahm J, Merboldt KD, Hanicke W. Direct flash MR imaging ofmagnetic field inhomogeneities by gradient compensation.MagnReson Med1988;6:474–80.

24. Yang QX, Dardzinski BJ, Li S, et al. Multi-gradient echo withsusceptibility inhomogeneity compensation (MGESIC): demon-stration of fMRI in the olfactory cortex at 3.0T.Magn Reson Med1997;37:331–5.

25. Constable RT. Functional MRI using gradient echo EPI in thepresence of large static field inhomogeneities.J Magn ResonImaging1995;5:746–52.

26. Constable RT, Spencer DD. Composite image formation inz-shimmed functional MRI.Magn Reson. Med.1999;42:110–7.

27. Constable RT, Carpentier A, Pugh K, et al. Investigation of thehippocampal formation using a randomized event-related para-digm and z-shimmed functional MRI.Neuroimage2000;12:55–62.

28. Glover G. 3D z-shim method for reduction of susceptibility ef-fects in BOLD fMRI. Magn Reson Med1999;42:290–9.

29. Frahm J, Merboldt K-D, Hänicke W. Functional MRI of humanbrain at high spatial resolution.Magn Reson Med1993;29:139–44.



30. Young IR, Cox IJ, Bryant DJ, et al. The benefits of increasingspatial resolution as a means of reducing artifacts due to fieldinhomogeneities.Magn Reson Med1988;6:585–90.

31. Merboldt KD, Finterbusch J, Frahm J. Reducing inhomogeneityartifacts in functional MRI of human brain activation: thin sec-tions vs. gradient compensation. J Magn Reson2000;145:184–91.

32. Penfield W, Boldrey E. Somatic motor and sensory representationin the cerebral cortex of man as studied by electrical stimulation.Brain 1937;60:389–443.

33. Kido DK, LeMay M, Levinson AW, et al. Computed tomographiclocalization of the precentral sulcus.Radiology1980;190:85–92.

34. Sobel DF, Gallen CC, Schwartz BJ, et al. Locating the centralsulcus: comparison of MR anatomic and magnetoencephalo-graphic functional methods.AJNR Am J Neuroradiol1993;14:915–26.

35. Naiditch TP, Valavanis AG, Kubik S. Anatomic relationshipsalong the low-middle convexity: Part I: normal specimens andmagnetic resonance imaging.Neurosurgery1995;36:517–32.

36. Uematsu S, Lesser R, Fisher RS, et al. Motor and sensory cortexin humans: topography studied with chronic subdural stimulation.Neurosurgery1992;31:59–72.

37. Sabatini U, Toni D, Pantano P, et al. Motor recovery after earlybrain damage: a case of brain plasticity.Stroke1994;25:514–7.

38. Wood C, Spencer D, Gibson W. Localization of human sensori-motor cortex during surgery by cortical surface recordings ofsomatosensory-evoked potentials. J Neurosurg1988;68:99–111.

39. Woolsey C, Erikson T, Gibson W. Localization of somatic sen-sory and motor areas of human sensory cortex as determined bydirect recording of evoked potentials and electrical stimulation.J.Neurosurg.1979;51:476–506.

40. Berger MS, Kincais J, Ojemann GA. Brain mapping techniques tomaximize resection, safety and seizure control in children withbrain tumors.Neurosurgery1989;25:786–92.

41. Schlosser MJ, McCarthy G, Fulbright RK, et al. Cerebral vascularmalformations adjacent to sensorimotor and visual cortex: func-tional magnetic resonance imaging studies before and after thera-peutic intervention.Stroke1997;28:1130–7.

42. Jack CR, Thompson RM, Butts RK, et al. Sensory motor cortex:correlation of presurgical mapping with functional MR imagingand invasive cortical mapping.Radiology1994;190:85–92.

43. Pardo FS, Aronen HJ, Kennedy D, et al. Functional cerebralimaging in the evaluation and radiotherapeutic treatment planningof patients with malignant glioma.Int J Radiat Oncol1994;30:663–9.

44. Puce A, Constable RT, Luby ML, et al. Functional magneticresonance imaging of sensory and motor cortex: comparison withelectrophysiological localization.J Neurosurg1995;83:262–70.

45. Yousry TA, Schmid UD, Jassoy AG, et al. Topography of thecortical motor hand area: prospective study with functional MRimaging and direct motor mapping at surgery.Radiology1995;195:23–9.

46. Righini A, de Divitiis O, Prinster A, et al. Functional MRI: pri-mary motor cortex localization in patients with brain tumors.JComput Assist Tomogr1996;20:702–8.

47. Schad LR, Bock M, Baudendistel K, et al. Improved target vol-ume definition in radiosurgery of arteriovenous malformations bystereotactic correlation of MRA, MRI, blood bolus tagging, andfunctional MRI.Eur Radiol1996;6:38–45.

48. Atlas SW, Howard RS, Maldjian J, et al. Functional magneticresonance imaging of regional brain activity in patients with in-tracerebral gliomas: findings and implications for clinical man-agement.Neurosurgery1996;38:329–38.

49. Chapman PH, Buchbinder BR, Cosgrove GR, et al. Functionalmagnetic resonance imaging for cortical mapping in pediatricneurosurgery. Pediatr Neurosurg1996;23:122–6.

50. Kahn T, Schwabe B, Bettag M, et al. Mapping of the corticalmotor hand area with functional MR imaging and MR imaging-guided laser-induced interstitial thermotherapy of brain tumors:work in progress.Radiology1996;200:149–57.

51. Maldjian J, Atlas SW, Howard RS, et al. Functional magneticresonance imaging of regional brain activity in patients with in-tracerebral arteriovenous malformations before surgical or endo-vascular therapy.J Neurosurg1996;84:477–83.

52. Mueller WM, Yetkin FZ, Hammeke TA, et al. Functional mag-netic resonance imaging mapping of the motor cortex in patientswith cerebral tumors.Neurosurgery1996;39:515–20.

53. Pujol J, Conesa G, Deus J, et al. Clinical application of functionalmagnetic resonance imaging in presurgical identification of thecentral sulcus. J Neurosurg1998;88:863–9.

54. Achten E, Jackson GD, Cameron JA, et al. Presurgical evaluationof the motor hand area with fMRI in patients with tumors anddysplastic lesions.Radiology1999;210:529–38.

55. Rao SM, Binder JR, Bandettini PA, et al. Functional magneticresonance imaging of complex human movements.Neurology1993;43:2311–8.

56. Rao SM, Binder JR, Hammeke TA, et al. Somatotopic mappingof the human primary motor cortex with functional magneticresonance imaging.Neurology1995;45:919–24.

57. Maldjian JA, Gottschalk A, Patel RS, et al. The sensory somato-topic map of the human hand demonstrated at 4 Tesla.Neuroim-age1999;11:473–81.

58. Lotze M, Erb M, Flor H, et al. fMRI evaluation of somatotopicrepresentation in human primary motor cortex.Neuroimage2000;11:473–81.

59. Beatse E, Sunaert S, Wilms GE, et al. The combined study oflanguage and sensory motor regions in patients with brain lesions:a functional MRI study using a mixed blocked and event relatedparadigm.Radiology2000;217(suppl):203.

60. Gold S, Christian B, Arndt S, et al. Functional MRI statisticalsoftware packages: a comparative analysis.Hum Brain Map1998;6:73–84.

61. Hermann BP, Perrine K, Chelune GJ, et al. Visual confrontationnaming following left anterior temporal lobectomy: a comparisonof surgical approaches.Neuropsychology1999;13:3–9.

62. Binder JR, Price CJ. Functional imaging of language. In: CabezaR, Kingstone A, eds.Handbook of functional neuroimaging ofcognition.Cambridge, MA: MIT Press, 2001:187–251.

63. Shallice T.From neuropsychology to mental structure.Cam-bridge, UK: Cambridge University Press, 1989.

64. Wise R, Chollet F, Hadar U, et al. Distribution of cortical neuralnetworks involved in word comprehension and word retrieval.Brain 1991;114:1803–17.

65. Mazoyer BM, Tzourio N, Frak V, et al. The cortical representa-tion of speech.J Cogn Neurosci1993;5:467–79.

66. Price CJ, Wise RJS, Warburton EA, et al. Hearing and saying: thefunctional neuro-anatomy of auditory word processing.Brain1996;119:919–31.

67. Binder JR, Frost JA, Hammeke TA, et al. Human temporal lobeactivation by speech and nonspeech sounds.Cereb Cortex2000;10:512–28.

68. Henschen SE. On the hearing sphere.Acta Otolaryngol1918;1:423–86.

69. Lehéricy S, Cohen L, Bazin B, et al. Functional MR evaluation oftemporal and frontal language dominance compared with theWada test.Neurology2000;54:1625–33.

70. Binder JR, Frost JA, Hammeke TA, et al. Conceptual processingduring the conscious resting state: a functional MRI study.J CognNeurosci1999;11:80–93.

71. Benson RR, FitzGerald DB, LeSeuer LL, et al. Language domi-nance determined by whole brain functional MRI in patients withbrain lesions.Neurology1999;52:798–809.

72. Petersen SE, Fox PT, Posner MI, et al. Positron emission tomo-graphic studies of the cortical anatomy of single-word processing.Nature1988;331:585–9.

73. Raichle ME, Fiez JA, Videen TO, et al. Practice-related changesin human brain functional anatomy during nonmotor learning.Cereb Cortex1994;4:8–26.

74. Warburton E, Wise RJS, Price CJ, et al. Noun and verb retrievalby normal subjects: studies with PET.Brain 1996;119:159–79.

75. Pardo JV, Fox PT. Preoperative assessment of the cerebral hemi-spheric dominance for language with CBF PET.Hum Brain Map1993;1:57–68.

76. Bahn MM, Lin W, Silbergeld DL, et al. Localization of languagecortices by functional MR imaging compared with intracarotidamobarbital hemispheric sedation.Am J Radiol1997;169:575–9.

77. Yetkin FZ, Swanson S, Fischer M, et al. Functional MR of frontal

FUNCTIONAL MRI 61


lobe activation: comparison with Wada language results.AJNRAm J Neuroradiol1998;19:1095–8.

78. Binder JR, Swanson SJ, Hammeke TA, et al. Determination oflanguage dominance using functional MRI: a comparison withthe Wada test.Neurology1996;46:978–84.

79. Binder JR, Frost JA, Hammeke TA, et al. Human brain languageareas identified by functional MRI.J Neurosci1997;17:353–62.

80. Demb JB, Desmond JE, Wagner AD, et al. Semantic encodingand retrieval in the left inferior prefrontal cortex: a functionalMRI study of task difficulty and process specificity.J Neurosci1995;15:5870–8.

81. Desmond JE, Sum JM, Wagner AD, et al. Functional MRI mea-surement of language lateralization in Wada-tested patients.Brain 1995;118:1411–9.

82. Fitzgerald DB, Cosgrove GR, Ronner S, et al. Location of lan-guage in the cortex: a comparison between functional MR imag-ing and electrocortical stimulation. AJNR Am J Neuroradiol1997;18:1529–39.

83. Yetkin FZ, Mueller WM, Morris GL, et al. Functional MR acti-vation correlated with intraoperative cortical mapping.AJNR AmJ Neuroradiol1997;18:1311–5.

84. Loring DW, Meador KJ, Lee GP, et al. Wada memory perfor-mance predicts seizure outcome following anterior temporal lo-bectomy.Neurology1994;44:2322–4.

85. Sperling MR, Saykin AJ, Glosser G, et al. Predictors of outcomeafter anterior temporal lobectomy: the intracarotid amobarbitaltest.Neurology1994;44:2325–30.

86. Weinand ME, Carter LP. Surface cortical cerebral blood flowmonitoring and single photon emission computed tomography:prognostic factors for selecting temporal lobectomy candidates.Seizure1994;3:55–9.

87. Manno EM, Sperling MR, Ding X, et al. Predictors of outcomeafter anterior temporal lobectomy: positron emission tomography.Neurology1994;44:2331–6.

88. Wada J, Rasmussen T. Intracarotid injection of sodium amytal forthe lateralization of cerebral speech dominance.J Neurosurg1960;17:266–82.

89. Milner B, Branch C, Rasmussen T. Study of short-term memoryafter intracarotid injection of sodium amytal.Trans Am NeurolAssoc1962;87:224–6.

90. Detre JA, Maccotta L, King D, et al. Functional MRI lateraliza-tion of memory in temporal lobe epilepsy.Neurology1998;50:926–32.

91. Killgore WDS, Glosser G, Casasanto D, et al. Functional MRIand the Wada test provide complementary information for pre-dicting post-operative seizure control.Seizure2000;8:450–5.

92. Binder JR, Bellgowan PSF, Swanson SJ, et al. fMRI activationasymmetry predicts side of seizure focus in temporal lobe epi-lepsy.Neuroimage2000;11:155.

93. Casasanto DJ, Glosser G, Killgore WDS, et al. Presurgical fMRIpredicts memory outcome following anterior temporal lobec-tomy. J Int Neuropsychol Soc2001;7:183.

94. Squire LR. Memory and the hippocampus: a synthesis from find-ings with rats, monkeys, and humans.Psychol Rev1992;99:195–231.

95. Stern CE, Corkin S, González RG, et al. The hippocampal for-mation participates in novel picture encoding: evidence fromfunctional magnetic resonance imaging.Proc Natl Acad Sci U SA 1996;93:8660–5.

96. Glosser G, Saykin AJ, Deutch GK, et al. Neural organization ofmaterial-specific memory functions in temporal lobe epilepsy pa-tients as assessed by the intracarotid amobarbital test.Neuropsy-chology1995;9:449–56.

97. Wagner AD, Schacter DL, Rotte M, et al. Building memories:remembering and forgetting of verbal experiences as predicted bybrain activity.Science1998;281:1188–91.

98. Mizrahi E, Kelleway P, Grossman R, et al. Anterior temporallobectomy and medically refractory temporal lobe epilepsy ofchildhood.Epilepsia1990;31:302–12.

99. Rasmussen T, Milner B. The role of early left-brain injury indetermining lateralization of cerebral speech functions.Ann N YAcad Sci1977;299:355–69.

100. Ojemann G, Ojemann J, Lettich E, et al. Cortical language local-

ization in left, dominant hemisphere: an electrical stimulationmapping investigation in 117 patients.J Neurosurg1989;71:316–26.

101. Devinsky O, Perrine K, Llinas R, et al. Anterior temporal lan-guage areas in patients with early onset temporal lobe epilepsy.Ann Neurol1993;34:727–32.

102. Springer JA, Binder JR, Hammeke TA, et al. Language domi-nance in neurologically normal and epilepsy subjects: a functionalMRI study.Brain 1999;122:2033–45.

103. Bookheimer SY, Zeffiro T, Blaxton T, et al. A direct comparisonof PET activation and electrocortical stimulation mapping forlanguage localization. Neurology1997;48:1056–65.

104. Disbrow EA, Slutsky DA, Roberts TP, et al. Functional MRI at1.5 Tesla: a comparison of the blood oxygenation level-dependentsignal and electrophysiology.Proc Natl Acad Sci U S A2000;97:9718–23.

105. Hertz-Pannier L, Gaillard WD, Mott S, et al. Noninvasive assess-ment of language dominance in children and adolescents withfunctional MRI: a preliminary study.Neurology 1997;48:1003–12.

106. Bookheimer SY, Dapretto M, Karmarkar U. Functional MRI inchildren with epilepsy.Dev Neurosci1999;21:191–9.

107. Logan WJ. Functional magnetic resonance imaging in children.Semin Pediatr Neurol1999;6:78–86.

108. Gaillard WD, Grandin CB, Xu B. Developmental aspects of pe-diatric fMRI: considerations for image acquisition, analysis, andinterpretation.Neuroimage2001;13:239–49.

109. Gaillard WD, Pugliesse M, Grandin CB, et al. Cortical localiza-tion of reading in normal children: a fMRI language study.Neu-rology 2001;57:47–54.

110. Bookheimer SY. Methodological issues in pediatric neuroimag-ing. Dev Med Ment Retard Rev2000;6:161–5.

111. Stapleton SR, Kiriakipoulos E, Mikulis D, et al. Combined utilityof functional MRI, cortical mapping, and frameless stereotaxy inthe resection of lesions in eloquent areas of brain in children.Pediatr Neurosurg1997;26:68–82.

112. Gaillard WD, Hertz-Pannier L, Mott SH, et al. Functionalanatomy of cognitive development: fMRI of verbal fluency inchildren and adults.Neurology2000;54:180–5.

113. Gaillard WD, Bookheimer SY, Cohen M. The use of fMRI inneocortical epilepsy. In: Williamson PD, Siegel AM, ThadaniVM, et al., eds.Advances in neurology: neocortical epilepsy.New York: Lippincott Williams & Wilkins, 2000:391–404.

114. Grandin CB, Gaillard WD, Petrella JR, et al. Comparison ofactivated brain areas during verbal fluency tasks in adults andchildren: an fMRI study.Proc Am Soc Neuroradiol1999;38:154.

115. Thomas KM, King SW, Franzen PL, et al. A developmentalfunctional MRI study of spatial working memory.Neuroimage1999;10:327–38.

116. Gaillard WD, Webb LK, Xu B, et al. Language neural networksof reading in young children identified with fMRI. Ann Neurol2000;48:517.

117. Balsamo LM, Xu B, Grandin CB, et al. Left hemisphere languagedominance with an auditory naming task in children: an fMRIstudy.Arch Neurol(in press).

118. Grandin CB, Gaillard WD, Whitnah JR, et al. Comparison ofphonological and semantic verbal fluency tasks: an fMRI study.Neuroimage1998;7:S133.

119. Keene DL, Logan WJ, McAndrews MP, et al. A comparison ofthree functional MRI language paradigms in children.Epilepsia2000;41(suppl 7):193.

120. Benson RR, Logan WJ, Cosgrove GR, et al. Functional MRIlocalization of language in a 9-year-old child.Can J Neurol Sci1996;23:213–9.

121. Gaillard WD, Theodore WH. Mapping language in epilepsy withfunctional neuroimaging.Neuroscientist2000;6:391–401.

122. Gaillard WD, Grandin CB, Pugliese M, et al. Identification oflanguage dominance in CPS patients using a fMRI reading task.Epilepsia1999;40(suppl 7):251.

123. Gaillard WD, Xu B, Balsamo L, et al. fMRI identification oflanguage dominance in patients with complex partial epilepsy



using an auditory based language comprehension task.Epilepsia2000;41(suppl 7):83.

124. Harvey AS, Anderson D, Jackson G, et al. Functional MRI lat-eralization of expressive language in children with partial epi-lepsy and left hemisphere lesions.Epilepsia 1999;40(suppl 7):183.

125. Westerveld M, Stoddard K, McCarthy K, et al. Case report offalse lateralization using fMRI: comparison of fMRI language

localization, Wada testing, and cortical stimulation.Arch ClinNeuropsychol1999;14:162–3.

126. Chelune GC. Hippocampal adequacy versus functional reserve:predicting memory functions following temporal lobectomy.Arch Clin Neuropsychol1995;10:413–32.

127. Loring DW, Meador KJ, Lee GP, et al. Wada memory asymme-tries predict verbal memory decline after anterior temporal lobec-tomy. Neurology1995;45:1329–33.

FUNCTIONAL MRI 63


Documents

Functional MRI in Epilepsy