EEG-fMRI of Focal Epileptic Spikes: Analysis With Multiple Haemodynamic Functions and Comparison With Gadolinium-Enhanced MR Angiograms

Human Brain Mapping 22:179 –192(2004) EEG-fMRI of Focal Epileptic Spikes: Analysis With Multiple Haemodynamic Functions and Comparison With Gadolinium-Enhanced MR Angiograms Andrew P. Bagshaw,* Yahya Aghakhani, Christian-G. Benar, ´ Eliane Kobayashi, Colin Hawco, Francois Dubeau, G. Bruce Pike, ¸ and Jean Gotman Montreal Neurological Institute, McGill University, Montreal, Quebec, Canada ´ ´ Abstract: Combined EEG-fMRI has recently been used to explore the BOLD responses to interictal epileptiform discharges. This study examines whether misspecification of the form of the haemodynamic response function (HRF) results in significant fMRI responses being missed in the statistical analysis. EEG-fMRI data from 31 patients with focal epilepsy were analysed with four HRFs peaking from 3 to 9 sec after each interictal event, in addition to a standard HRF that peaked after 5.4 sec. In four patients, fMRI responses were correlated with gadolinium-enhanced MR angiograms and with EEG data from intracranial electrodes. In an attempt to understand the absence of BOLD responses in a significant group of patients, the degree of signal loss occurring as a result of magnetic field inhomogeneities was compared with the detected fMRI responses in ten patients with temporal lobe spikes. Using multiple HRFs resulted in an increased percentage of data sets with significant fMRI activations, from 45% when using the standard HRF alone, to 62.5%. The standard HRF was good at detecting positive BOLD responses, but less appropriate for negative BOLD responses, the majority of which were more accurately modelled by an HRF that peaked later than the standard. Co-registration of statistical maps with gadolinium-enhanced MRIs suggested that the detected fMRI responses were not in general related to large veins. Signal loss in the temporal lobes seemed to be an important factor in 7 of 12 patients who did not show fMRI activations with any of the HRFs. Hum. Brain Mapp. 22:179 –192, 2004. © 2004 Wiley-Liss, Inc. Key words: event-related fMRI; EEG; epilepsy INTRODUCTION The simultaneous measurement of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) has recently been shown to provide valuable information concerning the localisation of the regions generating interictal epileptiform activity [Al Asmi et al., 2003; Jager et ¨ al., 2002; Krakow et al., 1999; Lazeyras et al., 2000; Lemieux et al., 2001]. A consistent feature of such studies is that a relatively low percentage of the patients who have interictal discharges whilst in the scanner demonstrate significant activations following statistical processing of the fMRI data. The reason for this is unclear. A possible explanation is that subtle changes in BOLD signal occur as a result of interictal Contract grant sponsor: Canadian Institutes of Health Research; Contract grant number: MOP 38079. *Correspondence to: Andrew P. Bagshaw, Montreal Neurological Institute, 3801 University Street Room 028, Montreal, QC, H3A 2B4, ´ Canada. E-mail: andrew.bagshaw@mcgill.ca Received for publication 2 September 2003; Accepted 20 January 2004 DOI 10.1002/hbm.20024 © 2004 Wiley-Liss, Inc. Bagshaw et al. discharges but the method by which the data is analysed is not detecting them. When the analysis is performed using a specific form for the haemodynamic response function (HRF), this could be a result of the time course of the BOLD changes being different than that assumed in the statistical processing. This would lead to a low correlation between the data and the model, despite the fact that a change in BOLD signal had occurred. There is considerable scope for misrepresenting the HRF during the statistical analysis, since there is a large variability in the measured BOLD response both within and between subjects, even for robust experimental activation paradigms such as motor and visual tasks [Aguirre et al., 1998; Duann et al., 2002; Miezin et al., 2000]. A further confounding factor that has not been addressed to any great extent is the question of whether even more variation might be expected from pathological, epileptic brain tissue. Certainly, the nature of the haemodynamic response to interictal spikes has not been fully characterised, although some preliminary results have been presented [Benar et al., ´ 2002; Lemieux et al., 2001]. The current study applied a range of different HRF models to the analysis of fMRI data collected interictally from patients with focal epilepsy. The models took the form of single gamma functions peaking 3, 5, 7, or 9 sec after each spike. The purpose was to see if the fMRI data contained significant deviations from baseline that were time-locked to the EEG activity but occurred with a different time course to that assumed in the original analysis using a standard HRF. Such an approach has been employed previously by Buckner et al. [1998] by delaying the onset of a single gamma function HRF by 2, 4, and 6 sec. The primary motivation behind the work was to determine whether an exploratory approach based on using several HRF models with different latencies could locate areas of activation that were previously undetected. The calculation of t maps retains information about the accuracy with which the modelled HRF accords with the actual BOLD response to the event. In addition, it is immediately apparent whether the measured response is an increase or decrease in the BOLD signal. This information is useful when considering the physiological interpretation of the BOLD response to interictal events. In particular, the origin of deactivations, or stimulus-locked decreases in the BOLD signal with respect to baseline, is currently a matter of considerable investigation [Born et al., 2002; Fransson et al., 1999; Hamzei et al., 2002; Harel et al., 2002]. A second point that was addressed was the origin of any new, and in particular delayed, areas of fMRI response in relation to the size of the blood vessels in the immediate vicinity. It has been suggested that delayed BOLD responses correlate with haemodynamic changes from large veins rather than venules [Krings et al., 1999; Lee et al., 1995], although some studies have observed considerable overlap in the temporal delays of the two groups [Hlustık et al., 1998; ˇ´ Saad et al., 2001]. Finally, it is known that echo-planar imaging (EPI) data demonstrate significant signal loss close to the boundaries between tissues that have very different magnetic susceptibilities, such as in the anterior temporal and orbito-frontal regions [Gorno-Tempini et al., 2002; Merboldt et al., 2000; Ojemann et al., 1997; Veltman et al., 2000]. Clearly, this will reduce the likelihood of detecting fMRI activations in patients whose epileptic spiking originates from these parts of the brain. Data from patients with spikes originating purely from the temporal lobes were analysed and the degree of signal loss compared with the presence or absence of fMRI activations. PATIENTS AND METHODS Patients Thirty-one patients were selected from a total population of 42 individuals with a clinical diagnosis of focal epilepsy and frequent spikes on routine EEG who underwent continuous EEG-fMRI monitoring (Table I). Only those imaging sessions that came from patients who had more than five interictal discharges during the course of the fMRI scanning period were included, leading to the exclusion of 11 patients. Written informed consent was obtained from all patients in accordance with the regulations of the Research Ethics Board of the Montreal Neurological Institute and Hospital. Three of the selected patients were scanned twice (data set numbers 1–19, 9 –10, and 12–18), resulting in 34 scanning sessions. Epileptic discharges were classified into different types by an experienced electroencephalographer according to their spatial distribution and morphology. Each type was analysed separately to allow for different BOLD responses. Five patients had more than one type of epileptic event (data set numbers 3– 4, 15–17, 22–23, 28 –29, and 37–38). In total, therefore, 40 data sets were analysed. Table I contains a brief description of the patients’ clinical status. The data sets are ordered such that the first 18 showed significant activations when analysed with the standard HRF. In four patients, double-dose gadolinium enhanced MR angiograms performed as part of a presurgical evaluation were compared with the fMRI results. Moreover, intracerebral EEG recordings were also used to correlate the source of the interictal events with the fMRI activations. In the remaining 27 patients, knowledge of the source of the interictal discharges was obtained from scalp EEG and clinical observations. Data Acquisition Functional MRI images were acquired in one of two 1.5T MR scanners (Vision and Sonata, Siemens, Erlangen, Germany) using an EPI sequence (voxel dimensions 5 5 5 mm, 25 slices, 64 64 matrix, TE 50 msec, TR 3 sec, flip angle 90 degrees). The fMRI data were collected in runs of 120 repetitions followed by a brief pause in the scanning. Each run took approximately 6 min, and between 6 and 12 runs were acquired per patient (see Table I). Twenty-one channels of EEG, arranged according to the International 10 –20 system [Binnie et al., 1982], were recorded simultaneously using an EMR32 amplifier (Schwarzer, Munich, Germany). The patient’s head was immobilized with a plastic bag filled with small polystyrene spheres, in which a vacuum was obtained by air suction (S&S 180 EEG-fMRI of Focal Epileptic Spikes TABLE I. Summary of the patients and results for each data set* Peak responses Activation Data set 1* 2 3 Clinical notes L TLE. L T calcification and gliosis posterior T. Bi T-O epilepsy. PNH. F-C focus, R MTS. Interictal scalp EEG LT L O, bi O Bi F fMRI runs (n) 8 10 10 Spikes (n) 7 129 14 HRF P5(P5) P7/P9 (P5/P7) G(G) Location LT LF Bi F Deactivation HRF P9(P9) P3(P5) P9(P9) Location RF Bi O RP Other responses Activation location R T-P, L F-C R F, L T, R T Bi thalamus, R T, pons, R O, cerebellum, P-C, F-C, L T Cerebellum, L T, R F, L T n/a Deactivation location L F, L F-C Corpus Callosum, pons, R C-P R F, C-P, L P 4 5 A As above R T neocortex and lower central (opercular) epilepsy. R C-T atrophy. L hemispheric focus with max T. No lesion. L T-O focus. Bi PNH. L C-F focus with max SMA. Middle cerebral artery porencephalic cyst. O-T lobe epilepsy R L. No lesion. O-T lobe epilepsy R L. No lesion R F epilepsy. No lesion. Bi F epilepsy, no focal abnormality. R T arachnoid cyst. Extensive R C-P focus. R hemispheric polymicrogyria. L T focus. Bi perisylvian polymicrogyria. L post quadrant focus. L P cortical dysplasia. As above RF R T, R C-P 9 54 33 G(G) P5(P5) Bi F R post T P9(P9) G(G) L caudate RP L P, R P, L C, L F n/a 6 7 8 B LT L post T Bi F-C-T max F, sometimes LF Bi O and post T, R L Bi O and post T, R L. RF Bi F Bi diffuse max F, R L LT R F-C 7 10 12 34 97 40 P7(G) P3/P5(P5) P3(P3) LT LT L F-C P9(P9) P9(P9) P7(P7) LP Corpus Callosum L F-C R P, post C, L T, R T, cerebellum LP Cerebellum, R F-T, R P, L T-O, R O n/a n/a R P-O, R F-T L T-P, R F-C, R F, L T, cerebellum L F-C, R C-P, L T, L P R P, L F, R T R O, cerebellum, L F, L O, L C, R P-T, R T, L T R T, C, L P, cerebellum, R F-C, R P, L T, L F, L F-C, L O Cerebellum, R T-O, R O, L F, R F, R T, L T L O, C-P, R T R P, L T n/a n/a L T, L F Bi thalamus, periventricular R F, L P, R O 9C 10C 11 12 D 7 7 9 6 9 10 9 20 90 245 15 8 21 9 G(G) n/a G(G) P5(G) P5(P5) P9(P7) P7/P9(G) LP n/a RF LF R C-P LT RF P7(P5) P7(P5) P9(P9) P9(P7) P7(P9) n/a P9(P9) R T-O R T-O C-P LT RP n/a LP L T-O R P, L O, L, F-P R thalamus, R F R T, bi thalamus, L P, R P LF n/a n/a 13D 14 15 16 L F-C 36 P7(G) RO n/a n/a n/a 17 18D 19* 20 As above Bi F epilepsy, no focal abnormality, R T arachnoid cyst. L TLE. L T calcification and gliosis post T. R T neocortex and lower central epilepsy. Postencephalitic, no lesion. Post-traumatic bi F epilepsy. Orbitofrontal atrophy and gliosis. Bi F-C Bi F LT RT 8 9 11 70 12 19 18 G/P7(G) G(G) P9(P9) n/a L T-O LF LT n/a P9(P9) n/a P7(P7) n/a RO n/a LT n/a n/a n/a n/a n/a 21 Bi F, L R 9 27 P3(P3) LF P7(P7) RF LT R P-O, L O, C, L F 181 Bagshaw et al. TABLE I. (continued) Peak responses Activation Data set 22 Clinical notes Multifocal epilepsy R and L F-T, R and L T-P. Postencephalitic, no lesion. As above L TLE. R post T focus. R hemispheric atrophy. L TLE. L oligodendroglioma. Bilateral O epilepsy. Bi T epilepsy. As above L MTS. Post-traumatic F epilepsy. L P focus. L focal cortical dysplasia. BL F epilepsy. BL F polymicrogyria. L C-P focus, Bi P-O polymicrogyria. Bi F-C epilepsy. No lesion. R T-O epilepsy. No lesion. Late onset bi TLE. No lesion. As above L MTS. L MTS. Interictal scalp EEG R F-T fMRI runs (n) 11 Spikes (n) 44 HRF n/a Location n/a Deactivation HRF P3(P3) Location LO Other responses Activation location n/a Deactivation location n/a 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 L F-T LT R post T RT R T-O LT RT LT LT L C-P L F-T Bi F Bi F-C R post T RT LT LT LT 10 10 10 11 12 10 6 8 10 10 6 10 9 205 29 16 24 162 21 7 10 6 56 175 9 8 14 60 13 10 85 n/a n/a P9(P7) n/a n/a n/a n/a n/a n/a P7(P7) n/a P3(P3) n/a n/a n/a n/a n/a n/a n/a n/a LO n/a n/a n/a n/a n/a n/a Post C n/a LP n/a n/a n/a n/a n/a n/a n/a n/a P5(P3) n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a P3(P3) n/a n/a n/a n/a n/a RF n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a LF n/a n/a n/a n/a n/a L T, L C n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a R P, L C, cerebellum, FC, L F n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a RO n/a n/a n/a 8 9 *For each data set, regions of activation and deactivation with the highest t value are given as the “Peak Responses.” The location is given, along with the HRF that gave the highest t value and, in parentheses, the maximum volume. An entry such as P5/P7 indicates that the statistical maps generated using P5 and P7 gave values that were the same. The locations of secondary regions of significant fMRI response are given under “Other Responses.” The data sets are ordered such that the first 18 showed significant responses in the original analysis with the standard HRF. R: right; L: left; F: frontal; T: temporal; P: parietal; O: occipital; C: central; post: posterior; bi: bilateral; G: Glover HRF. P3, P5, P7, P9 as in the text. n/a: no activation; TLE: temporal lobe epilepsy; PNH: periventricular nodular heterotopia; MTS: mesial temporal sclerosis. , ,} Patients with two sessions. ABCD denotes patients who had SEEG recording. X-Ray Products, Brooklyn, NY). Patients stayed in the magnet for approximately 90 min and seldom complained of discomfort. The EEG was filtered offline to remove the artefact generated by the MR scanner (FEMR software, Schwarzer) [Hoffmann et al., 2000]. During the same imaging session, an anatomical MRI was performed (1-mm slice thickness, 256 256 matrix, TE 9.2 msec, TR 22 msec, flip angle 30 degrees). Double-dose gadolinium enhanced MRIs, which are sensitive to the vasculature, including the venous phase of the circulation, were performed in four patients during a separate scanning session. The same parameters as for the anatomical scan were used. Image Analysis The fMRI images were motion corrected and smoothed with a Gaussian kernel (full width at half maximum 6 mm) using in-house software. For all patients, the maximum translation applied by the motion correction algorithm was less than 1.5 mm, and the rotation less than 1 degree. Statistical analysis of the functional MR images was performed using the methods and software of Worsley et al. [1996, 2002]. Differences in the slice acquisition time were corrected for. The first three frames of each run were not included in the analysis to ensure that the magnetisation was in a steady state. Models and signals were prewhitened with an auto-regressive filter of order 1, and low-frequency drifts in the fMRI signal were modelled with a third-order polynomial fitted to each run. The initial analysis assumed a haemodynamic response to interictal epileptiform discharges that was the same as that to brief auditory stimuli, which peaks after 5.4 sec (the standard or Glover HRF) [Glover, 1999]. Further ex- 182 EEG-fMRI of Focal Epileptic Spikes coregistered with the anatomical scan taken during the same session as the functional scans using in-house software. Since the functional images, and hence the statistical maps, were also registered to the T1-weighted anatomical scan, this has the effect of coregistering the gadolinium MRI with the statistical maps and thus allowing the position of activated areas to be compared with the veins visible on the gadolinium MRI. However, even if the EPI images have been motion corrected, the coregistration of EPI and T1 weighted data can have errors of several millimetres due to differences in the geometric distortions between the two methods [Studholme et al., 2000]. In order to account for possible misregisration, therefore, a functional MRI activation was considered to be unconnected with a large vein if the distance between the two was greater than 5 mm, or one voxel of the EPI data. In this context, large veins were classified as those with clear signal enhancement in the gadolinium MRI, the “macrovasculature” defined by Krings et al. [1999]. Figure 1. The five HRFs used in the analysis. P3 to P9 are single gamma functions with the same width as the standard (Glover) HRF. ploratory analyses were performed using a modelled HRF that was composed of a single gamma function with the same width as the first gamma function used in the standard model (5.2 sec FWHM), but with a time to peak of 3, 5, 7, and 9 sec (P3, P5, P7, and P9). The second gamma function of the standard HRF was not included in order to reduce the number of variable parameters and maintain the overall shape of the HRF. The five HRFs are presented in Figure 1. Significant activation was defined as five contiguous voxels above a t value of 3, which corresponds to a P value of 0.01 [Cao, 1999]. The level of significance of the composite analysis is therefore approximately 0.05 as a result of the fact that each data set is being analysed five times, once for each of the five HRFs. The statistical threshold was chosen to account for the fact that analysing each data set several times increases the likelihood of false activations. For a single analysis, a P value of 0.05 would correspond to three contiguous voxels above a t value of 3. Reducing the P value for an individual analysis to 0.01, or five voxels above a t value of 3, maintains the overall significance level for the multiple analyses at 0.05. Using a lower threshold would increase the number of false activations and thus confuse the effects of analysing with five HRFs. An fMRI activation was considered to be consistent with the EEG activity if both were in the same lobe. Responses with positive t values were labelled as activations and those with negative t values were labelled as deactivations. For each of the four patients who underwent EEG recording from intracerebral electrodes (stereotaxic EEG [SEEG]), the gadolinium-enhanced MRIs were Details of Scalp EEG and SEEG in Implanted Patients Four patients had previously undergone SEEG recording according to the methods of Olivier et al. [1996] as part of a presurgical evaluation for intractable epilepsy that could not be localised with scalp EEG. The patients will be labelled A (data set 5), B (data set 8), C (data sets 9 and 10), and D (data set 13). Table II gives details of the electrode placement for these patients, as well as details of their interictal scalp EEG, and interictal and ictal SEEG. Estimation of Signal Loss in Patients With Temporal Spiking In order to focus specifically on those patients in whom fMRI signal loss as a result of susceptibility artefacts would be expected to be important, ten patients were selected whose spikes originated from the temporal lobes. Only those patients were included whose spikes were limited to electrodes T3/T4, T5/T6, F7/F8, or Fp1/Fp2. An assessor (EK) who was blinded to the patient’s identity and the results of the fMRI analysis qualitatively estimated the level of signal loss in the temporal lobes using categories 1 (minimal signal loss) to 4 (severe signal loss). This was done by overlaying the functional EPI data onto the anatomical T1 scan for each patient in order to locate the temporal lobes in three orthogonal planes. The level of signal loss was then estimated by thresholding the functional image at a level of 10% of the maximum signal in the EPI image and assessing the percentage of voxels below that threshold. In defining the categories, the signal in the whole of the temporal lobes was considered, with particular attention to the cortex and mesial structures, and to the homogeneity of the signal. Figure 2 gives examples of data from each category. For each category, a single transverse slice through the temporal 183 Bagshaw et al. TABLE II. Summary of the scalp EEG, SEEG, and fMRI data collected in four patients Patient A Scalp EEG; interictal 1) R T 2) R C-P 3) Rare R F-C SEEG; electrode placement Bilateral implantation; 5 Depth on R: Am, anterior Hc, posterior Hc, orbilofrontal, and cingulate gyrus. 2 Depth on left: Hc, and orbitofrontal. 11 Epidural: R first T gyrus, Superior and inferior pre and post C gyri. L unilateral implantation 4 Depth; Am, Hc, SMA, cingulate 12 Epidural: first temporal and precentral gyri, and around the lesion SEEG; interictal 1) R inferior C 2) R Am, Hc and T neocortex 3) R pHc and posterior T neocortex SEEG: ictal onset 15 clinical seizures; R T neocortex and adjacent suprasylvian area Several pure EEG seizures; R Am & Hc plus neocortex fMRI responses Positive (P5): R posterior T - inferior P (max t 7.0, 59 voxels) Negative (G): R Pparasagittal (max t 3.8, 13 voxels) fMRI/EEG agreement fMRI is congruent with the most active epileptiform activity and a C-T neocortical lesion. B 1) Bilateral F-C-T, max F 2) L F 1) L SMA and cingulate 2) L pre C 3) L post C 24 clinical seizures; synchronous in L SMA, cingulate, and precentral cortex C Bilateral O and posterior T, R L D Bilateral diffuse, R L, max F Bilateral implantation; 9 Depth: bilateral Hc, PHc, Inferior P, Superior O, and R Inferior O. 9 Epidural: first temporal, angular, supramarginal gyri, and L interior O R unilateral implantation Depth: None 11 Epidural: inferior pre and post C gyrus, mid pre and post C gyrus, frontal opercular region, orbitofrontal, supramarginal and angular gyri, posterior P, P-O, and posterior T 1) Bilateral mesial O, R L 2) Bilateral Hc, L R 3) Bilateral posterior T-O neocortex, L R 8 clinical seizures; Bilateral synchronous in O and posterior T neocortex Positive (P3): cingulate gyrus connected to parasagittal. Also L dorso-lateral F (max t 7.9, 281 voxels in total) Negative (P7): L F convexity (max t 6.5, 132 voxels) and L posterior P parasagittal (max t 5.5, 42 voxels) Negative (P7): R basal T and O and posterior P (max t 8.7, 42 voxels); L posterior basal T and O (max t 6.4, 16 voxels) fMRI is congruent with the maximum epileptogenic area. Lesion was extensive and involved the F-C-P on the L. fMRI is congruent with the widespread epileptogenic area. No visible lesion. 1) R angular gyrus 2) R frontal operculum 3) R central 5 clinical seizures: 3 seizures from pre and post central regions. 2 seizures from post central region Positive (P5): R P in parietal polymicrogyric cortex. Activation also connected to mesial central region (max t 9, 232 voxels in total). Negative (P7): R P (max t 5.4, 7 voxels) fMRI is congruent with the lesion (R perisylvian polymicrogyria) but less congruent with the widespread eplleptogenic area. R: right; L: left; F: frontal; T: temporal; P: parietal; O: occipital; C: central; Am: amygdala; Hc: hippocampus, SMA: supplementary motor area; PHc: parahippocampus. lobes is shown, with the functional data superimposed onto the corresponding anatomical slice. As far as possible, equivalent slices were selected for each patient, although differences in the slice orientation mean that exact correspondence was not always possible. RESULTS Analysis With Multiple HRFs The 34 imaging sessions in 31 patients resulted in 40 data sets, taking into account the fact that some patients had more than one type of interictal epileptiform discharge in a particular session. Table I contains the details of the analysis, along with a brief description of the clinical diagnoses of the patients. For each data set, the regions of activation and deactivation with the highest t values were chosen. Table I gives the HRF with which the t value and, in parentheses, the volume were maximum. For example, the activation from session 1 had both a maximum t value and a maximum volume when the data were analysed with P5. Similarly, the deactivation had a maximum t value and a maximum volume when the data were analysed with P9. An entry in Table I such as P5/P7 indicates that the statistical maps generated using P5 and P7 gave values that were the same. Figure 3 shows examples of the effect on the detected fMRI responses of using the optimum HRF, compared with the standard analysis. A number of data sets showed more than one region of significant activation and deactivation. The locations of these regions are included in Table I as “Other Responses,” but only the main clusters of activation and deactivation were analysed further. The data sets are ordered in Table I such that the first 18 showed significant activations when analysed with the stan- 184 EEG-fMRI of Focal Epileptic Spikes Figure 2. Examples of the functional data in each signal loss category. A single transverse slice is shown, with the functional MRI data superimposed on the anatomical scan. dard HRF. Any activation for data sets 19 to 40 was thus not apparent with the standard analysis. Such activations were found in seven data sets, raising the percentage with significant activation from 45% with the standard HRF alone to 62.5% when using multiple HRFs. The new activations agreed with the known EEG and clinical data concerning the localisation of the epileptic focus in three of the seven data sets. Of the 40 data sets, 5 (12.5%) showed only activations, 3 (7.5%) showed only deactivations, 17 (42.5%) showed both activations and deactivations, and 15 (37.5%) did not show a response with any of the modelled HRFs. The areas of deactivation never overlapped spatially with an activation found using an HRF that peaked earlier, or vice versa (a late negative response coinciding spatially with an earlier positive response could be interpreted as the undershoot of the positive response). Figure 4 summarises the data from Table I concerning the timing of the fMRI responses. Only those data sets are included for which a particular HRF gave results that were clearly superior (i.e., a clearly increased t value or cluster volume) to those obtained with the other HRFs. Data sets 2, 7, 15, and 17 are excluded from Figure 4 because no HRF gave clearly superior results. As can be seen from Figure 4, there was a tendency for the activations to be most significant, particularly in terms of their spatial extent, when using the standard HRF, for which 10 out of 22 were maximum. Other than that, the maximum Figure 3. Examples of differences in fMRI responses between the standard and the optimum HRF analyses. A,C,E: The standard analysis for a particular fMRI response. B,D,F: The optimum analysis for a particular fMRI response. Standard (A) and P9 (B) analysis of data set 14. Standard (C) and P7 (D) analysis of data set 6 (activation). Standard (E) and P9 (F) analysis of data set 6 (deactivation). Increases in the volume and maximum t statistic can be observed. 185 Bagshaw et al. Figure 4. The HRF model for which the t value and volume were maximum is plotted for all data sets that had a clearly superior value with one particular HRF (see Table I). Deactivations show a tendency to peak later than activations. activations are distributed relatively evenly between the different HRFs. This is in contrast to the deactivations, which show a tendency to peak later. Nine of the 20 data sets that had deactivations showed a maximum t value when using P9. Fifteen of the 20 had a maximum negative t value when using P7 or P9, i.e., an HRF that peaked later than the standard HRF (5.4 sec). Twelve patients did not show any activation despite having a sufficient number of spikes to be included in the study (8 to 175 spikes in the session). There does not appear to be a correlation between the distribution of spikes, the type of epilepsy, or the absence or presence of lesions and the probability of finding significant fMRI responses (Table I). The results from the three patients who underwent two scanning sessions were variable. In the first (data sets 9 and 10), there was excellent agreement between the two sessions, both spatially and in terms of the HRF for which the deactivations were maximum. There was also good agreement between the fMRI and the EEG data, both from the scalp and intracerebral electrodes (see patient C below). The results for the other two patients were less consistent. For one (data sets 1 and 19), the first scanning session led to an area of activation that was consistent with the EEG and maximum with P5. This was not apparent with P5 in the second session, but a similar region was activated with P9 (Fig. 5). In the final patient (data sets 12 and 18), both sessions demonstrated regions of activation, but only those in the first session were consistent with the EEG and clinical data. Figure 5. Comparison of standard and optimum analyses of two scans in the same patient (data sets 1 and 19). Standard (A) and P5 (B) analysis of data set 1. Standard (C) and P9 (D) analysis of data set 19. Similar regions are activated in the two sessions, but with different HRFs. 186 EEG-fMRI of Focal Epileptic Spikes Correlation Between Venous Drainage, fMRI Activations, and SEEG Four patients had SEEG and gadolinium-enhanced MRIs for evaluation of venous drainage. Details are contained in Table II. polymicrogyria), but less so with the widespread epileptogenic region defined by the SEEG. Correlation Between Signal Loss and fMRI Activations Table III summarises the results of comparing the severity of signal loss with the presence or absence of fMRI activation in the temporal lobes. Of the ten patients who fulfilled the selection criterion (spikes limited to the temporal lobes), three demonstrated significant fMRI responses with at least one HRF (data sets 5, 6, and 19). These were also the only three patients for whom the signal loss was estimated as category 1 (minimal signal loss). The remaining seven patients, or a total of nine data sets, did not show fMRI responses with any of the HRFs. The signal loss in these seven patients varied between categories 2 and 4. Patient A. A significant activation and a significant deactivation were detected in the right posterior temporal-inferior parietal region. Neither of these were related to large veins. The area of deactivation was superior to the area of activation, and less significant than it. Both the activation and deactivation were in the same lobe as the most active interictal epileptiform activity measured on the scalp EEG and SEEG, as well as being consistent with a right centro-temporal lesion and the origin of the clinical seizures on SEEG. Patient B. This patient had an fMRI activation in the cingulate gyrus that was connected to activation in the parasagittal and left dorso-lateral frontal regions (Fig. 6A,B). Parts of the region of activation were approximately 5 mm from large veins (the sagittal sinus and the internal cerebral vein), but were consistent with the interictal epileptiform activity observed using implanted electrodes (the supplementary motor area and cingulate gyrus). Of the two deactivations (Fig. 6C,D), one was close to the epileptogenic region defined by the SEEG, but the other was quite removed from it. The deactivation in the posterior parietal region is approximately 5 mm from the sagittal sinus, and may be related to it. DISCUSSION The simultaneous measurement of EEG and fMRI is capable of providing valuable information concerning the sources of epileptic spikes. However, if the technique is to realise its potential and become a routine part of presurgical evaluation, it must be capable of accurately and reliably locating the regions that are responsible for the epileptic events. In the current study, the clusters with the highest positive and negative t values were investigated in each patient. It will always be the case when dealing with a heterogeneous group of patients, many of whom have widespread or multifocal EEG abnormalities, that several regions will show significant fMRI responses. If combined EEGfMRI is ever to be used in a clinical setting, it will have to provide useful information in a wide range of patients, and it is likely that the most attention will be given to the region that shows the strongest correlation with the scalp EEG spikes. In an analogous manner, if a patient undergoing EEG monitoring has frequent right temporal spikes but occasional left temporal spikes, the conclusion may be drawn that the right temporal region is most abnormal, whilst acknowledging that the left temporal region is also not strictly normal. At the present stage of the development of EEG-fMRI, therefore, it is most important to understand the correlation between the main regions of fMRI activation and deactivation and the EEG. Undoubtedly, the secondary regions contain useful and interesting information that will help to understand the nature of the link between scalp EEG spikes and the fMRI response in epilepsy. One obstacle to using EEG-fMRI in a clinical setting is the fact that when studies involving relatively large groups of heterogeneous patients have been reported, a significant proportion of the patients have not had fMRI activations. In the few previously reported studies that have examined ten or more patients with focal epilepsy, the percentage of patients who have shown significant fMRI responses varied between 39 and 70%. Krakow et al. [1999] examined ten patients with frequent spikes (an average of 1 per minute) and very focal patterns of epileptic activity on the EEG, and Patient C. Bilateral deactivations with right-sided predominance were observed in the basal temporal and occipital regions, consistent with the interictal scalp EEG and the widespread epileptogenic area as defined by the SEEG. The right occipital deactivation was not near a large vein, but the left occipital deactivation was within 5 mm of the lateral sinus. It was much larger than the vein, even considering the effect of the low spatial resolution of the functional images, which is likely to spread the BOLD response away from the vessel into the parenchyma [Lai et al., 1993]. These deactivations are shown in Figure 7. Patient D. A large region of activation was detected, predominantly in the right parietal region in an area of polymicrogyric cortex but connected with the mesial central region. It was not close to any large veins. The interictal SEEG recordings showed epileptiform activity in the right angular gyrus, the right frontal operculum, and the right central region. Five clinical seizures recorded on the SEEG originated from the pre- and post-central regions. The fMRI activation was, therefore, congruent with the lesion (right perisylvian 187 Bagshaw et al. Figure 6. A: activation in the region of the supplementary motor area from response on the gadolinium-enhanced MRI. Parts of the regions data set 8 (patient B) overlaid on the gadolinium enhanced MRI. shown in A and B are close to large veins, but are consistent with The statistical map was generated using P3. B: A cingulate gyrus SEEG data. The deactivation shown in C is close to the epileptoactivation in the same session, same patient, also with P3. C: A genic region defined by the scalp EEG and SEEG and is not close deactivation using P7. D: Another deactivation using P7. The to any veins, while that in D is not consistent with the epileptoellipses in the left panels show the approximate extent of the fMRI genic region and is close to the sagittal sinus. observed significant fMRI responses in seven of them, although the response in one of these patients was not reproducible in a subsequent examination (see below). In a further study, Krakow et al. [2001] examined 24 patients with the same inclusion criteria, and detected significant fMRI responses in 14 of them (58%), two of which did not agree with the electrographic and clinical findings. Al Asmi et al. [2003] had less stringent inclusion criteria, examining 38 patients with a wide range of spike frequencies and those with multi-focal discharges, similarly to the current study, and observed fMRI activation in 39% of them. None of these studies specifically looked at potential reasons for the lack of activation in some patients. In the current work, the form of the modelled HRF and the influence of susceptibility-related 188 EEG-fMRI of Focal Epileptic Spikes Figure 7. A: Deactivation in the right temporal region from scan 9 in patient the gadolinium enhanced MRI. Both regions are consistent with C overlaid on the gadolinium enhanced MRI. B: A left temporal the epileptogenic zone defined by the scalp EEG and SEEG. The deactivation in the same session, same patient. Both statistical deactivation shown in B is within 5 mm of a large vein, but is maps were generated using P7. The ellipses in the left-side images considerably larger than it. in A and B show the approximate extent of the fMRI response on TABLE III. Results of a qualitative assessment of the degree of signal loss in the temporal lobes in ten patients who had spikes limited to the temporal lobes* Data set number 5 6 19 20 26 28 29 30 37 38 39 40 Spike distribution F8, T4 T3, F7 T3, F7, Fp1 F8, T4 F8, T4 F7, T3, FP1, T5 F8, T4, FP2, T6 F7, T3 T4, T6, F8 T3, F7, T5 Same amplitude T3, F7 Same amplitude T3, F7 Signal loss category 1 1 1 3 4 4 3 4 2 2 fMRI activation in T Lobe? Y Y Y N N N N N N N signal losses have been examined in order to assess the role they play in determining whether fMRI responses will be detected. The effect of the modelled HRF was examined by using HRFs based on single gamma functions in order to retain as much information as possible concerning the timing of the responses. The use of a more accurate HRF will increase the likelihood of detecting voxels where the signal-to-noise ratio is low, finding new areas of BOLD response and increasing the volume of responses that had previously been detected, thus giving a more realistic estimate of the extent of the response [Saad et al., 2003]. In addition, by using an HRF that corresponds more closely to the actual BOLD response, the estimated cluster P value will better reflect the true probability of activation, rather than a composite value that combines the probability of activation and the accuracy of the model. Several other methods exist for dealing with inter-patient variability in the shape of the HRF, such as using a more flexible set of basis functions [Friston et al., 1998] or calculating HRFs based on the data for each patient *The electrodes involved in the spiking are arranged in descending order of spike amplitude, unless otherwise stated. 189 Bagshaw et al. [Kang et al., 2003; Marrelec et al., 2003]. Using several HRFs with different latencies is complementary to these and has the combined advantages of a specific shape to the HRF, hence a potential advantage in terms of the matched filter theorem, and some flexibility in the type of response that can be detected. The use of multiple HRFs increased the percentage of data sets with activation from 45% to 62.5% when compared with the analysis with a standard HRF. It should be noted that this increase is not caused by the repeated analyses since the statistical threshold was changed to take this into account. In fact, the statistical threshold that has been used is quite conservative. Despite this, in four data sets the only significant fMRI responses that were detected were not consistent with the EEG and clinical data concerning the location of the source of the spikes. The interpretation of such responses remains to be investigated. The results suggest that the standard model of the haemodynamic response is reasonably good at detecting positive BOLD responses to interictal epileptiform events. Most activations were apparent when using the standard model, and the standard model gave the highest t value and volume most often. The standard HRF seems to be less appropriate, however, for detecting regions of negative BOLD signal change. The interpretation of stimulus-locked negative BOLD signal changes, or deactivations, is still under investigation [Born et al., 2002; Fransson et al., 1999; Hamzei et al., 2002; Harel et al., 2002]. Although an increase in the local deoxyhaemoglobin concentration is considered to be responsible for the decrease in BOLD signal, the cause of this increase is not clear. A combination of a reduction in neuronal activity and haemodynamic changes independent of the local changes in neuronal activity (i.e., vascular steal) has been proposed [Shmuel et al., 2002]. In the present study, regions of negative BOLD signal change were frequently not accompanied by associated regions of positive BOLD signal change, which makes it unlikely that they are caused by vascular steal mechanisms [Shulman et al., 1997]. The use of additional techniques to measure blood flow and oxygenation [e.g., Schulte et al., 2001; Strangman et al., 2002; Toronov et al., 2003] would help to shed light on the mechanisms that lead to the frequent deactivations seen in patients with epilepsy. Deactivations show a clear tendency to occur later, but the reason for this is not clear. Some studies have suggested that activations from large draining vessels tend to occur later than those from smaller veins [Krings et al., 1999; Lee et al., 1995], but the results of coregistering the gadolinium-enhanced MRIs with the statistical maps suggest that in general neither the activations nor deactivations, whether prompt or delayed, were related to large veins. The fact that the positive and negative responses in a particular patient do not overlap spatially demonstrates that the negative responses are not just the undershoot of an earlier positive response, but are stimulus-locked primary decreases in the BOLD signal. One of the problems encountered in EEG-fMRI in epilepsy is that it is often difficult to validate the fMRI activations. The comparison of SEEG data and fMRI activations is one method of accomplishing this, although SEEG is only available in patients who are being considered for surgical intervention but whose epilepsy can not be well localised from scalp EEG recordings. SEEG electrodes are placed in likely epileptogenic regions as determined by scalp EEG, imaging, and clinical manifestations, but they cover a limited portion of the cortex, and thus offer only a limited comparison with the results of fMRI studies. However, SEEG data is often one of the most important tools available to the clinician, and thus one of the most important points of comparison for EEG-fMRI activations. Of the 31 patients studied, four had undergone SEEG recording. In three of them the scalp and intracerebral EEG defined the same epileptogenic area, and the fMRI responses were entirely consistent with this. In the other patient (patient D), the epileptogenic region defined by both sets of electrophysiological recordings was more widespread than the fMRI response, but close to the lesion presumably causing the epilepsy. It is unclear whether this is a result of the fMRI detecting the focus of the epileptic activity while the scalp EEG and SEEG are sensitive to other regions of propagation. The three patients who underwent two scanning sessions demonstrate that the repeatability of EEG-fMRI studies in epilepsy is a question that needs to be considered more closely. In only one of the patients were the results from the two sessions fully consistent. In the other two, either the timing of the haemodynamic response or the location of the activation was different between the two sessions. Symms et al. [1999] addressed this issue by scanning a single patient with focal spikes four times and found reproducible changes in all of them. Krakow et al. [1999] defined significant activations as those that were consistent in repeated studies, and detected them in six of ten patients. Another patient had activation in only one of two experiments. Both of these studies required the patients to have very frequent spikes and excluded those with multifocal spiking patterns. Clearly, a systematic study of patients with focal and multifocal spiking would help to determine the factors that affect the reproducibility of EEG-fMRI studies in epilepsy. Twelve patients did not show fMRI activations with any of the five HRFs. Although there is no clear difference in the clinical diagnoses of these patients compared with others who did show activation, 11 of the 12 had either a temporal or frontal component to the distribution of their interictal epileptiform discharges on the EEG. This raises the question of whether signal loss due to magnetic field inhomogeneities might be preventing the detection of BOLD signal changes in these regions. The analysis of the EPI data from ten patients with temporal lobe spikes suggests that this is the case. It is striking that the only patients in this group who showed fMRI activations were those in whom the signal loss was estimated to be minimal (i.e., category 1). It is, thus, likely that 7 of the 12 patients in the study who did not have any fMRI responses had a temporal lobe generator that was 190 EEG-fMRI of Focal Epileptic Spikes not detected because of signal loss in the temporal lobes. In addition, three other patients without significant fMRI responses had frontal lobe epilepsy, which is the other main region affected by signal loss caused by susceptibility artefacts. This highlights the importance of implementing methods to try to reduce the effect of susceptibility related signal loss when attempting to use EEG-fMRI to locate the sources of interictal spikes. Several methods have been proposed. Reducing the echo time from the 50 ms that was used in the present work is a technically straightforward solution that will help to recover signal in the affected areas, but perhaps not sufficiently to allow BOLD signal changes in those regions to be detected [Gorno-Tempini et al., 2002]. It will also reduce the BOLD sensitivity throughout the brain. Reduction of the slice thickness is also relatively easy to implement, but work by Merboldt et al. [2000] suggests that 1-mm slices are optimum, which would be difficult with 2D acquisition and will prevent the whole of the brain from being scanned unless a prohibitively long repetition time is used. In EEG-fMRI of epileptic patients, it is preferable to scan the whole brain, since the generator of scalp spikes may, within reason, be anywhere. Other strategies include optimisation of the slice orientation [Deichmann et al., 2003; Ojemann et al., 1997], the use of diamagnetic materials to correct for field perturbations [Wilson et al., 2002], and zshimming methods, which apply refocusing pulses to correct for local field inhomogeneities in the slice selection direction [e.g., Cordes et al., 2000, Gu et al., 2002]. All of these techniques have drawbacks, such as precluding whole brain coverage, reducing the temporal resolution, reducing the BOLD contrast to noise ratio, or, in the case of passive diamagnetic shims, discomfort for the patient during the course of a 90-min scanning session. In conclusion, it seems that misspecification of the form of the HRF has an important impact on the probability of detecting significant BOLD responses in epileptic patients. A considerable increase in the percentage of patients in whom significant BOLD responses were detected was noted when using multiple HRFs in the analysis. In addition, the use of multiple HRFs highlighted different trends in the timing of activations and deactivations, with deactivations tending to occur later than activations. A relatively high percentage of patients who had spikes in the scanner still did not have any significant activations or deactivations, despite analysis with five HRFs. The data from patients with temporal lobe spikes strongly suggests that this may be due to signal loss as a result of magnetic field inhomogeneities. ous and spike-triggered EEG-fMRI studies of epileptic spikes. Epilepsia 44:1328 –1339. Aguirre GK, Zarahn E, D’Esposito (1998): The variability of human BOLD hemodynamic responses. Neuroimage 8:360 –369. Benar C-G, Gross DW, Wang Y, Petre V, Pike B, Dubeau F, Gotman ´ J (2002): The BOLD response to interictal epileptiform discharges. Neuroimage 17:1182–1192. Binnie CD, Rowan AJ, Gutter T (1982): A manual of electroencephalographic technology. Cambridge, UK: Cambridge University Press. Born AP, Law I, Lund TE, Rostrup E, Hanson LG, Wildschiødtz G, Lou HC, Paulson OB (2002): Cortical deactivation induced by visual stimulation in human slow wave sleep. Neuroimage 17:325–1335. Buckner RL, Koutstaal W, Schacter DL, Dale AM, Rotte M, Rosen BR (1998): Functional-anatomic study of episodic retrieval II Selective averaging of event-related fMRI trials to test the retrieval success hypothesis. Neuroimage 7:163–175. Cao J (1999): The size of the connected components of excursion sets of chi2, t, F fields. Adv Appl Prob 31:579 –595. Cordes D, Turski PA, Sorenson JA (2000): Compensation of susceptibility-induced signal loss in echo-planar imaging for functional applications. Magn Res Imag 18:1055–1068. Deichmann R, Gottfried JA, Hutton C, Turner R (2003): Optimized EPI for fMRI studies of the orbitofrontal cortex. Neuroimage 19:430– 441. Duann J-R, Jung T-P, Kuo W-J, Yeh T-C, Makeig S, Hsieh J-C, Sejnowski TJ (2002): Single-trial variability in event-related BOLD signals. Neuroimage 15:823– 835. Fransson P, Kruger G, Merboldt K-D, Frahm J (1999): MRI of func¨ tional deactivation: temporal and spatial characteristics of oxygenation-sensitive responses in human visual cortex. Neuroimage 9:611– 618. Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R (1998): Event-related fMRI: characterizing differential responses. Neuroimage 7:30 – 40. Glover GH (1999): Deconvolution of impulse response in eventrelated BOLD fMRI. Neuroimage 9: 416 – 429. Gorno-Tempini ML, Hutton C, Josephs O, Deichmann R, Price C, Turner R (2002): Echo time dependence of BOLD contrast and susceptibility artefacts. Neuroimage 15:136 –142. Gu H, Feng H, Zhan W, Xu S, Silbersweig DA, Stern, E, Yang Y (2002): Single-shot interleaved z-shim EPI with optimized compensation for signal losses due to susceptibility-induced field inhomogeneity at 3T. Neuroimage 17:1358 –1364. Hamzei F, Dettmers C, Rzanny R, Liepert J, Buchel C, Weiller C ¨ (2002): Reduction of excitability (“inhibition”) in the ipsilateral primary motor cortex is mirrored by fMRI signal decreases. Neuroimage 17:490 – 496. Harel N, Lee S-P, Nagaoka T, Kim D-S, Kim S-G (2002): Origin of negative blood oxygenation level dependent fMRI signals. J Cereb Blood Flow Metab 22:908 –917. Hlustık P, Noll DC, Small SL (1998): Suppression of vascular arteˇ´ facts in functional magnetic resonance images using MR angiograms. Neuroimage 7:224 –231. Hoffmann A, Jager L, Werhahn, KJ, Jaschke M, Noachter S, Reiser M ¨ (2000): Electroencephalography during functional echo-planar imaging: detection of epileptic spikes using post-processing methods. Magn Reson Med 44:791–798. Jager L, Werhahn KJ, Hoffmann A, Berthold S, Scholz V, Weber J, ¨ Noachtar S, Reiser M (2002): Focal epileptiform activity in the brain: detection with spike-related functional MR imaging: preliminary results. Radiology 223:860 – 869. ACKNOWLEDGMENTS C.G.B. was funded by a CIHR doctoral research award. E.K. was funded by a Preston Robb Fellowship from the Montreal Neurological Institute. REFERENCES Al Asmi A, Benar C-G, Gross DW, Agha Khani Y, Andermann F, ´ Pike B, Dubeau F, Gotman J (2003): fMRI activation in continu- 191 Bagshaw et al. Kang JK, Benar C-G, Al-Asmi A, Khani YA, Pike GB, Dubeau F, ´ Gotman J (2003): Using patient-specific hemodynamic response functions in combined EEG-fMRI studies in epilepsy. Neuroimage 20:1162–1170. Krakow K, Woermann FG, Symms MR, Allen PJ, Lemieux L, Barker GJ, Duncan JS, Fish DR (1999): EEG-triggered functional MRI of interictal epileptiform activity in patients with partial seizures. Brain 122:1679 –1688. Krakow K, Lemieux L, Messina D, Scott CA, Symms MR, Duncan JS, Fish DR (2001): Spatio-temporal imaging of focal interictal epileptiform activity using EEG-triggered functional MRI. Epilept Disord 3:67–74. Krings T, Eberich SG, Roessler F, Reul J, Thron A (1999): MR blood oxygenation level-dependent signal differences in parenchymal and large draining vessels: implications for functional MR imaging. Am J Neuroradiol 20:1907–1914. Lai S, Hopkins AL, Haacke EM, Li D, Wasserman BA, Buckley P, Friedman L, Meltzer H, Hedera P, Friedland R (1993): Identification of vascular structures as a major source of signal contrast in high resolution 2D and 3D functional activation imaging of the motor cortex at 1.5T: preliminary results. Magn Reson Med 30:387–392. Lazeyras F, Blanke O, Perrig S, Zimine I, Golay X, Delavelle J, Michel CM, de Tribolet N, Villemure J-G, Seeck M (2000): EEGtriggered functional MRI in patients with pharmacoresistant epilepsy. J Magn Res Imag 12:177–185. Lee AT, Glover GH, Meyer CH (1995): Discrimination of large venous vessels in time course spiral blood-oxygen-level-dependent magnetic resonance neuroimaging. Magn Reson Med 33:745–754. Lemieux L, Salek-Haddadi A, Josephs O, Allen P, Toms N, Scott C, Krakow K, Turner R, Fish DR (2001): Event-related fMRI with simultaneous and continuous EEG: description of the method and initial case report. Neuroimage 14:780 –787. Marrelec G, Benali H, Ciuciu P, Pelegrini-Issac M, Poline JB (2003): Robust Bayesian estimation of the hemodynamic response function in event-related BOLD fMRI using basic physiological information. Hum Brain Mapp 19:1–17. Merboldt K-D, Finsterbush J, Frahm J (2000): Reducing inhomogeneity artefacts in functional MRI of human brain activations: thin sections vs. gradient compensation. J Magn Res 145:184 –191. Miezin FM, Maccotta L, Ollinger JM, Petersen SE, Buckner RL (2000): Characterizing the hemodynamic response: effects of presentation rate, sampling procedure, and the possibility of ordering brain activity based on relative timing. Neuroimage 11:735–759. Ojemann JG, Akbudak E, Snyder AZ, McKinstry RC, Raichle ME, Conturo TE (1997): Anatomic localization and quantitative analysis of gradient refocused echo-planar fMRI susceptibility artefacts. Neuroimage 6:156 –167. Olivier A, Alonso-Vanegas M, Comeau R, Peters T (1996): Imageguided surgery of epilepsy. Neurosurg Cin N Am 7:229 –243. Saad ZS, Ropella KM, Cox RW, DeYoe EA (2001): Analysis and use of fMRI response delays. Hum Brain Mapp 13:74 –93. Saad ZS, Ropella KM, DeYoe EA, Bandettini PA (2003): The spatial extent of the BOLD response. Neuroimage 19:132–144. Schulte A-C, Speck O, Oesterle C, Hennig J (2001): Separation and quantification of perfusion and BOLD effects by simultaneous acquisition of functional I0 and T2* parameter maps. Magn Reson Med 45:811– 816. Shmuel A, Yacoub E, Pfeuffer J, Van de Moortele P-F, Adriany G, Hu X, Ugurbil K (2002): Sustained negative BOLD, blood flow and oxygen consumption response and its coupling to the positive response in the human brain. Neuron 36:1195–1210. Shulman GL, Fiex JA, Corbetta M, Buckner RL, Miezin FM, Raichle ME, Petersen SE (1997): Common blood flow changes across visual tasks: II. Decreases in cerebral cortex. J Cogn Neurosci 9:648– 663. Strangman G, Culver JP, Thompson JH, Boas DA (2002): A quantitative comparison of simultaneous BOLD fMRI and NIRS recordings during functional brain activation. Neuroimage 17: 719 –731. Studholme C, Constable RT, Duncan JS (2000): Accurate alignment of functional EPI data to anatomical MRI using a physics-based distortion model. IEEE Trans Med Im 19: 11 1115–1127. Symms MR, Allens PJ, Woermann FG, Polizzi G, Krakow K, Barker GJ, Fish DR, Duncan JS (1999): Reproducible localization of interictal epileptiform discharges using EEG-triggered fMRI. Phys Med Biol 44:N161–N168. Toronov V, Walker S, Gupta R, Choi JH, Gratton E, Hueber D, Webb A (2003): The roles of changes in deoxyhemoglobin concentration and regional cerebral blood volume in the fMRI BOLD signal. Neuroimage 19:1521–1531. Veltman DJ, Friston KJ, Sanders G, Price CJ (2000): Regionally specific sensitivity differences in fMRI and PET: where do they come from? Neuroimage 11:575–588. Wilson JL, Jenkinson M, Jezzard P (2002): Optimization of static field homogeneity in human brain using diamagnetic passive shims. Magn Reson Med 48:906 –914. Worsley KJ, Marrett S, Neelin P, Vandal AC, Friston KJ, Evans AC (1996): A unified statistical approach for determining significant signals in images of cerebral activation. Hum Brain Mapp 4:58–73. Worsley KJ, Liao CH, Aston J, Petre V, Duncan GH, Morales F, Evans AC (2002): A general statistical analysis for fMRI data. Neuroimage 15:1–15. 192