Electroencephalogram-triggered functional magnetic resonance imaging in focal epilepsy


Dr Senichiro Kikuchi, Department of Psychiatry, Jichi Medical School, 3311-1, Yakushiji, Minamika-wachi-machi, Kawachi-gun, Tochigi 329-0498, Japan. Email: skikuchi@jichi.ac.jp


Abstract  The high spatial resolution and cost performance of functional magnetic resonance imaging (fMRI) is useful for estimating focus localization in epilepsy, but this is difficult in the case of ictal fMRI because this is susceptible to motion artifacts. Electroencephalogram (EEG)-triggered fMRI, which is interictal, can be performed without marked movement and is thought to be useful, but requires further investigation in order to establish a methodology. The authors studied EEG-triggered fMRI in partial epileptic patients. Six patients were examined using a Nihon Kohden digital EEG recorder and Signa Horizon High Speed LX 1.5 T MRI scanner. Six electrodes were attached in the vicinity of the focus detected by scalp EEG. The fMRI scans were recorded after the discharges (activation) and scans without spikes (baseline). Equal numbers of activation and baseline scans were collected and analyzed using SPM99. In three of the six patients, an activated area was observed near the focus, but no activated areas were found in the other three subjects who tended to have a low number of spikes and low spike amplitude. Although various approaches focusing on improvement of the activation/non-activation ratio are required, EEG-triggered fMRI is a promising technique for detecting focal epileptic brain activity.


Various examination methods are needed to clarify the mechanism and diagnosis of epilepsy. Recently, trials of estimation of epileptic focus localization have been conducted using neuroimaging methods. There have been many studies of single-photon emission computed tomography (SPECT)1,2 and positron emission tomography (PET).3,4 But the spatial resolution of SPECT is relatively low among neuroimaging methods currently in general use, and PET scans are expensive.5 Functional magnetic resonance imaging (fMRI) may be a suitable method to examine hemodynamics, but it is difficult to detect seizure focus during the ictal period because of motion artifacts generated from marked movement such as general tonic clonic seizure.6 Functional MRI is especially susceptive to motion artifact, so its application is limited to estimating the hemodynamics of seizures with very slight movement.

Electroencephalogram (EEG)-triggered fMRI measures hemodynamics arising from spike discharge by using EEG and fMRI together. This method can detect interictal spike-related changes in hemodynamics that cannot be detected by SPECT or PET. Moreover, EEG-triggered fMRI can be performed without marked movement. There have been many studies of this application,7–9 but cases exist where the focus cannot be localized. Further investigation is needed to establish a methodology. We studied EEG-triggered fMRI on patients with partial epilepsy.



The present subjects were six partial epilepsy outpatients (age, 14–52 years; two men and four women) of the Neuropsychiatry Department of Gunma University Hospital (Table 1). Spike localization was produced in several scalp EEG recordings, in which focal spikes of all subjects were consistently found in the fixed channel. None of the subjects had abnormal neurological manifestations. We performed screening brain computed tomography (CT) and MRI on all of the patients to check for gross neoplastic or vascular lesions, and detected none. Informed consent was obtained from all subjects before this study was conducted.

Table 1.  Clinical, EEG and fMRI patient data
PatientAge, sexType of epilepsyInterictal EEGSample imagesSpike amplitudes (µV)
  1. EEG, electroencephalogram; fMRI, functional magnetic resonance imaging; FLE, frontal lobe epilepsy; OLE, occipital lobe epilepsy; TLE, temporal lobe epilepsy.

148 FFLERight frontal46164.70 ± 4.74
246 FOLERight occipital25148.72 ± 11.05
323 mTLERight temporal12 98.58 ± 5.82
452 mTLELeft temporal 7101.71 ± 9.67
514 FTLELeft temporal14133.79 ± 5.00
644 FTLELeft temporal 9121.78 ± 15.86

Electroencephalogram recorders and electrodes

We used a Nihon Kohden EEG4518 digital multichannel EEG-recorder (sampling frequency, 500 Hz; Nihon Kohden, Tokyo, Japan). The scalp EEG disk electrodes were made of Ag/AgCl (diamagnetic substances), and their diameter and the thickness were less than those of electrodes usually used for scalp EEG recordings in order to reduce their effect on the MR images. As one of the pilot studies for safety, we scanned a phantom consisting of a block of butter with electrodes installed, and found that the phantom never melted. Because six-channel non-magnetic electrodes were installed in the vicinity of the focus, electrode positioning varied depending on the subject. After attaching the electrodes, we covered the subject's head with a cotton net to prevent the cords from making loops capable of causing electric currents by electromagnetic induction. The electrodes were connected to an EEG recording system, which digitized and transmitted the EEG signal out of the scanner room through a diamagnetic cable and then reconstructed the analog EEG signals. We were unable to maintain the signal/noise (S/N) ratio of MR images when we passed the EEG cable through the open door of the magnet room, so we connected the cable to a 9-pin connector that is part of the radio frequency (RF) filter. During the study, none of the subjects had serious accidents (e.g. burning), and none had epileptic seizures.

Functional magnetic resonance imaging and data acquisition

A Signa Horizon High Speed LX MR system (General Electric, WI, USA) operating at 1.5 T and equipped for echo-planar imaging was used to acquire the anatomical and functional MR images. After attaching the electrodes, a foam-padded air cushion was positioned to restrict movement because the slightest movement of the subject's scalp, including the pulse of blood through it, would move the electrodes to an extent that is sufficient to generate large ballistocardiogram signals that may obscure the underlying brain activity. The subjects lay supine inside the magnet bore with head resting inside a circularly polarized quadrature head coil used for radiofrequency transmission and reception. The subject's nasion was aligned with a laser cross-hair for approximate head centering within the standing magnetic field. The parameters of the blood oxygenation level-dependent (BOLD) sequence were: repetition time/echo time (TR/TE) = 3000/40 ms; number of excitations (NEX) = 1; matrix = 64 × 64; field of view (FOV) = 220 × 220 mm; flip angle (F/A) = 90; 20 × 5 mm interleaved slices; no gap. These settings provided the best coverage with the least distortion on the same scanner. The parameters of the T1-weighted anatomical images were TR/TE = 440/9 ms, NEX = 2, matrix = 512 × 192, FOV = 220 × 220 mm, F/A = 90, slice thickness: 5 mm; no gap.

The on-line EEG signals were monitored by an expert, who marked the epileptic discharges. Images were acquired as ‘activation’ images and ‘baseline’ images. An ‘activation’ image was a single volume scan after a single stereotyped epileptic discharge. The peak time of BOLD signal after the epileptic discharge still remains unclear, but it is thought that the change in peak blood oxygenation level detected by fMRI occurs 4 s and 7 s after the onset of brief brain activity and return completely to baseline level at approximately 15 s.10 We started acquisition at between 2 s and 4 s after an EEG event with manual trigger in order to capture a single volume scan from 5 s to 7 s (time lag: 2–4 s; 1 TR = 3 s) after the spike, which was approximately the peak time after the brief brain activity. We acquired only a single volume scan per spike because it was thought that the first volume scan would allow a more accurate analysis than the second volume scan. ‘Baseline’ images were acquired after at least 20 s of background EEG activity without spike in order to avoid the contamination of the BOLD signal change according to the spike. All images were visually inspected and those with high T1 effect or heavy susceptibility artifact were excluded from analysis.

Data analysis

Equal  numbers  of  activation  and  baseline  images were used for statistical analysis. We processed the data using SPM99 statistical parametric mapping software11,12 (Wellcome Department of Cognitive Neurology, London, UK) on MATLAB (Mathworks, MA, USA). Realignment and smoothing (Gaussian Kernel, full width at half maximum (FWHM): 6 mm) were performed, and voxel-by-voxel analysis was carried out using a one-condition box-car design. ‘Activation’ images were condition 1, and ‘baseline’ images were condition 0. The threshold for significant activation was P < 0.05, corrected. We collected images but these were not time series so we switched off the hemodynamic response function (HRF) option to apply the box-car design, which was originally designed for time series.


We obtained 46, 25, 12, 7, 14 and nine activation images from patients 1–6 (Table 1), respectively, and an equal number of baseline images. An example of the EEG recordings of patient 5 at activation state is shown in Fig. 1. Three patients (1, 2 and 5) had activated areas near the focus as shown in Fig. 2(a–c), respectively. Activated areas were detected at the right frontal lobe of patient 1 (arrow in Fig. 2a), right occipital lobe of patient 2 (arrow in Fig. 2b), and left hippocampus of patient 5 (arrow in Fig. 2c). No activated areas were observed in patients 3, 4 or 6. The mean sample number of epileptic discharges was 28.33 ± 16.26 for patients 1, 2, and 5, and was 9.33 ± 2.52 for patients 3, 4, and 6. Table 1 shows the mean of spike amplitude for each patient, which was 154.91 ± 13.75 µV for patients 1, 2 and 5, and 106.82 ± 14.85 µV for patients 3, 4 and 6.

Figure 1.

Patient 5: electroencephalogram (EEG) recording inside the magnetic resonance (MR) scanner (bandpass filter frequency 0.3–60 Hz). Spike is clearly detected over the left temporal area (arrow). After 2 s of the spike, image acquisition artifact of echo planar imaging (EPI) acquisition occurs.

Figure 2.

(a) Patient 1, activated area, axial section (arrow at right frontal lobe); (b) patient 2, activated area, axial section (arrow at right occipital lobe); (c) patient 5, activated area, coronal section (arrow at left hippocampus).


Jackson et al. Krings et al. and Kubota et al. have examined epileptic regions by analyzing ictal and non-ictal T2*-weighted images (ictal fMRI).6,13,14 However, ictal fMRI is susceptible to motion artifacts and requires a high seizure frequency because maximum ictal fMRI scan time is approximately 20–30 min. The rarity of such cases greatly limits the applicability of ictal fMRI.

The focus is often estimated based on the spike discharge, although this may not always be the area of the epileptic seizure. It may be useful to estimate the location of the focus based on spike discharge, although this estimate may not always be the site of the epileptic seizure. EEG-triggered fMRI has recently been studied as a possible solution. We were able to image the activated area generated by the spike using current equipment, and consequently confirmed activated areas in three of the six subjects (50%).

In previous studies, Krakow confirmed activated areas in 12 of 24 cases (50%),8 and Patel et al. confirmed activated areas in 12 of 34 cases (35%);7 these ratios are not particularly high. The reason that the presumed foci areas failed to represent significant BOLD signal change as ‘activation’ is unknown, but may be related to the small number of epileptic discharges, given the low number of the spike and/or low amplitudes of discharges, because many images are required to allow sufficient averaging to detect the small BOLD signal changes for fMRI analysis. In the present study, more activation was found in subjects with a greater number of samples of epileptic discharge and/or spike amplitude. The EEG-triggered fMRI may be useful to estimate the location of the spike focus, especially in subjects with high spike count and/or a high spike amplitude.

During studies of EEG-triggered fMRI, care should be taken to avoid artifacts in EEG generated from MRI and, likewise, artifacts in MRI generated from EEG.15,16 We reduced EEG ballistocardiogram artifacts by using a foam-padded air cushion. To maintain the S/N ratio of MRI, we attached non-magnetic electrodes to the scalp that are smaller and thinner than those normally used in usual scalp EEG recordings, and connected the EEG cable via an RF filter.

To establish a methodology, various approaches are being investigated, especially techniques focused on improving the activation/non-activation ratio. Patel et al. reported that they were able to increase the ratio using novel methods of analysis.7 Salek-Haddadi et al. and Anami et al. performed simultaneous recording of EEG and fMRI by using digital processing to eliminate MR scanning noise.9,17

Although various approaches focused on improvement of the activation/non-activation ratio are required, EEG-triggered fMRI is a promising technique for detecting focal epileptic brain activity.


The present study was supported in part by grants from the Japan Epilepsy Research Foundation. We wish to thank Michiharu Suzuki and Takayuki Suto (Gunma University School of Medicine) for technical assistance with MRI, Keiichi Miyamoto (Real NeuroTechnology) for LAN support in MR data transfer, and Daisuke Koyama (Nihon Kohden) for advice on the use of EEG recorder and electrodes.