Introduction

Magnetoencephalographic (MEG) data have been used in determining the location of epileptic foci and eloquent brain function in patients with demonstrable lesion-related focal epilepsies.^{1, 2} Magnetic source imaging (MSI) involves the superimposition of locations of brain activity, measured by MEG, on to MR images;^3–5 MSI can delineate the somatosensory cortex for purposes of preoperative assessment.⁵ For presurgical epilepsy evaluation, MSI can locate the source of an epileptic spike as an equivalent current dipole (ECD) and predict the epileptic zone in patients with refractory epilepsy with lesions.^{1, 2, 6, 7} A single spike dipole from an MEG spike cannot determine the spatial extent of the epileptic zone because the model represents the center of activation by a point source rather than the area of activated cortex.⁸ Previous studies have characterized distributions of MEG spike dipoles with respect to results from intraoperative electrocorticography (ECoG) and intracranial EEG (icEEG) data.^{2, 7, 9} A single MEG cluster, the localization of which was within and extended from areas of MRI-defined cortical dysplasia and at margins of tumors or cystic lesions, correlated highly with that of the ictal onset zone.^{2, 7, 10} Further, failure to resect the brain region containing the MEG cluster (e.g., eloquent cortex) leads to postoperative seizure recurrence.¹

This chapter describes characterization of MEG spike dipoles and discusses the combination of other modalities in patients with MRI-negative focal epilepsy, and various types of algorithm for analyzing MEG data.

MEG predicts outcome following surgery for MRI-negative refractory epilepsy

The outcome following epilepsy surgery in patients with normal brain MRI depends on the case selection criteria and expertise of the epilepsy center. There is no accurate estimate of the prevalence of normal MRI findings in patients with refractory epilepsy who are potential surgical candidates. The probability of excellent postsurgical outcome following nonlesional surgery is uniformly lower than lesional surgery across many studies in the literature. In a recently published meta-analysis,¹¹ only 34–45% of patients with nonlesional MRI were seizure-free after epilepsy surgery. In patients with temporal lobe epilepsy, the odds of being seizure-free after surgery were 2.7 times higher in those with lesions. In patients with extratemporal epilepsy the odds were 2.9 higher in those with lesions.

Smith et al.¹² reported the extent of the MSI focus resection associated with surgical outcome in 20 patients with nonlesional extratemporal lobe epilepsy. Eight of ten patients became seizure-free when their nonlesional extratemporal MSI focus was extensively resected, versus one of ten when the focus was partially or totally unresected. Knowlton et al.¹³ showed that the positive predictive value of MSI was 90% and the negative predictive value was 44% for successful short-term surgical outcomes in patients with extratemporal and mesial/lateral temporal lobe epilepsies. The subjects included those with nonlocalizing MR abnormalities or ambiguous abnormalities, e.g., large, multiple, subtle, or questionable lesions. In one series of 22 children with normal or nonfocal MRI findings and resective surgery,¹⁴ 17 (77%) children achieved a good postsurgical outcome (defined as Engel class IIIA or better), which included eight (36%) seizure-free children. All children with postsurgical seizure freedom had an MEG cluster in the final resection area. Postsurgical seizure freedom was obtained in none of the children who had bilateral MEG dipole clusters or only scattered dipoles. They concluded that the presence of a MEG dipole cluster confined to the resection area was a prerequisite for postsurgical seizure freedom, and the absence of the MEG dipole cluster, the presence of bilateral MEG dipole clusters, or scatters may not indicate the precise epileptic network to resect.

The MEG was able to differentiate two epileptic foci as independent clusters of MEG spike dipoles within the right frontocentral region in an MRI-negative adolescent with simple partial seizure with secondary generalization.¹⁵ Distribution of MEG spike dipoles have been classified according to their number and density in neocortical epilepsy.⁷ Clusters consisted of six or more spike dipoles with < = 1 cm between adjacent dipoles, whereas scatters consisted of fewer than six spike dipoles regardless of the distance between dipoles, or spike dipoles with > 1 cm between dipoles regardless of the number of dipoles in a group (Figure 6.1). The extent of distribution of MEG dipoles can be classified into focal, regional, hemispheric, or multifocal. The density of MEG spike dipoles can be described as cluster or scatter. Both the distribution and the density of the MEG spike dipoles play a role in understanding epileptic networks. Oishi et al.⁹ reported similar results in 20 neocortical epilepsy patients, including four MRI-negative patients. They found that a single MEG cluster is predictive of a seizure-free outcome. Ten of 14 patients with single clusters became seizure-free after the resection of the ictal onset zone on icEEG, in comparison with one of six patients with multiple MEG spike clusters (p = 0.049). More single clusters (nine of 14) coincided with the ictal onset zone than multiple clusters (none of six) (p = 0.014). Conversely, association between postsurgical outcome and MRI findings was not significant; postsurgical seizure freedom occurred in eight of 13 patients with a single MRI lesion, one of four patients with no lesion, and two of three patients with multifocal lesions. Although the MEG single cluster has been a predictive factor for good postsurgical outcome, concordance between EEG and MEG localization is also essential to accurately localize the epileptic focus;^{9, 14} MEG data can also provide critical information in the placement of intracranial electrodes.¹⁶

Figure 6.1 Three-dimensional rendering MRI with equivalent current dipoles (ECD). A 39-year-old female had daily seizures consisting of jitteriness-like movements in the extremities and facial grimacing, lasting 5–10 seconds in duration. MRI did not show any abnormality. The regional clustered MEG spike dipoles (red squares with tails) are localized in the right frontotemporal regions. The regional scattered MEG spike dipoles are localized in the posterior temporal and central head regions. Red squares indicate locations of ECDs and tails represent moment of ECD. Yellow circle represents somatosensory evoked field. Green circle represents auditory evoked field with 1 kHz tone burst. After intracranial video-EEG monitoring, she underwent resection of frontal lobe and anterior part of lateral temporal region including ictal onset zone and active interictal zone. White line indicates the posterior border of the resection. The regional clustered MEG spike dipoles were completely resected. She has been seizure-free while taking antiepileptic medications for 15 months after surgery. Histopathological finding was reported as “normal.”

In cases where MEG shows scattered dipoles in neocortical epilepsy, various possibilities such as focal or multifocal epileptic networks may underlie this finding. There are very few papers reporting patients with scattered MEG dipoles in surgical series of neocortical epilepsy.^{7, 14} Postsurgical seizure freedom was obtained in no children who had normal or nonfocal MRI findings and only scattered dipoles.¹⁴ In two children who had a neocortical lesion and only scattered MEG dipoles, they achieved seizure freedom after cortical excisions and multiple subpial transections over the entire scattered dipoles area where icEEG colocalized the epileptic zones.⁷ When scattered MEG dipoles are found in neocortical and MRI-negative epilepsy, repeat MEG studies may help to show clustered MEG dipoles correlating with active epileptiform discharges on EEG. Further studies are needed to to understand the implications of scattered dipoles in MRI-lesional and MRI-negative epilepsy surgery patients.

Complementary use of MEG and other noninvasive modalities for localizing an epileptogenic zone

The findings of MEG are often compared with neuroimaging modalities, rather than with the electrophysiologic findings from scalp video-EEG recordings. In a subset of MRI-positive cases, MEG spike dipoles are expected to localize the epileptogenic zone, and the dipoles may asymmetrically surround the epileptogenic lesion.² In contrast, MEG spike dipoles in MRI-negative cases have no landmark of a lesion with which to associate. Therefore, peri-ictal SPECT and subsequent subtraction image coregistered to MRI (SISCOM), and 2-deoxy-2-(¹⁸F)fluoro-_D-glucose PET (FDG-PET) have been applied for comparison with the MEG results in MRI-negative cases.

Seo et al.¹⁷ compared lobar localizing values among MEG, SISCOM, and statistical parametric mapping (SPM) analysis of FDG-PET in MRI-negative pediatric epilepsy patients. In 14 children with MRI-negative surgery, 7 (50%) were seizure-free after surgery; MEG (79%, 11/14) and SISCOM (79%, 11/14) showed significantly greater lobar concordance with icEEG than SPM-PET (13%, 3/14) (p < 0.05).

Schneider et al.¹⁸ compared the localization value of SISCOM and MEG in MRI-negative focal epilepsy. They screened 54 MRI-negative surgical candidates and selected 14 consecutive patients (10–53 years) to compare MEG and SISCOM data with the icEEG-based resection area. Their determination of the MEG seizure localization was the result of an equivalent current dipole model that required a minimum of five MEG spikes for localization. The SISCOM data were transformed into a Z score using the mean and standard deviation of the differences in all brain voxels.¹⁹ They used a Z score of 2 for determining SISCOM-based seizure localization. The MSI and SISCOM data were compared with icEEG results with respect to four regional concordance categories as “sublobar,” “lobar,” “multilobar,” or “nonlocalizing.” Sublobar concordance of icEEG and MEG, and complete focus resection was found in five (36%) of the 14 patients: four (80%) became seizure-free. Sublobar icEEG-MSI concordance and complete focus resection significantly increased the chance of postoperative seizure freedom (p = 0.038). In contrast, only four of the six (67%) patients with sublobar concordant icEEG and SISCOM and complete focus resection became seizure-free (p = 0.138). They concluded that MEG was more advantageous compared to SISCOM in predicting seizure-free epilepsy surgery outcome when sublobar concordance with icEEG was observed in MRI-negative focal epilepsy.

Widjaja et al.²⁰ reported complimentary use of FDG-PET and MEG to improve detection of the epileptogenic zone in localization-related pediatric epilepsy with normal or subtle changes on MRI. They evaluated the sensitivity, specificity, positive, and negative predictive values of lobar localization of MEG, FDG-PET, FDG-PET+MEG (defined as both of two tests concordant with the icEEG-based resection area), and FDG-PET/MEG (defined as one or both tests concordant with the resection area) in 22 children who underwent cortical resection. Fourteen of 16 patients with a MEG cluster concordant with cortical resection achieved postoperative seizure freedom, whereas ten of 15 patients with FDG-PET concordant with cortical resection were seizure-free postoperatively; MEG had higher sensitivity and specificity (85.0% and 99.1%, respectively) relative to FDG-PET (65.0% and 95.4%, respectively). FDG-PET+MEG had reduced sensitivity but increased specificity (55.0% and 100%, respectively) relative to individual test. In contrast, FDG-PET/MEG had increased sensitivity but reduced specificity (95.0% and 93.5%, respectively. In MRI-negative epilepsy, concordant results of MEG and FDG-PET suggested the location of the discrete epileptogenic zone, with good probability of excellent seizure outcome when both abnormal foci are resected (Figures 6.1 and 6.2). The two investigations were complementary in the assessment of children with localization-related epilepsy, particularly when one test was nonlocalizing or nonconcordant.

Figure 6.2 FDG-PET coregistered on to axial MRI. FDG-PET shows hypometabolism in the right frontal and temporal lobes, corresponding to the regional clustered MEG spike dipoles in Figure 6.1.

In a subset of patients with MRI-negative focal epilepsy, results of MEG can prompt the re-evaluation of the MRI review to improve the identification of structural brain lesions on MRI. Funke and colleagues²¹ re-evaluated the MRI after MEG data became available in 29 patients with suspected neocortical epilepsy. In 7 (24%) patients, MEG-guided review of MRI led to recognition of clear, but previously unidentified, lesions; MEG can be a useful adjunct to MRI for the identification of structural lesions in MRI-negative focal epilepsy. Currently available advanced techniques and modalities of MRI have become more relevant for detecting subtle lesions adjacent to the MEG spike dipoles in cases with previously diagnosed MRI-negative epilepsy.

Who are the patients most likely to benefit from MEG?

The MEG studies in extratemporal neocortical epilepsy had higher yields than did studies in patients with mesial temporal lobe epilepsy.^{14, 22}

In patients with frontal lobe epilepsy, the presurgical evaluation, and even the diagnosis, may be compounded by poor electroclinical localization, due to the rapid propagation of the EEG epileptiform and seizure discharges. Ossenblok and colleagues²³ have analyzed the epileptogenic localizing value of interictal MEG and simultaneous scalp EEG spikes by systematically using automated analysis procedure in 24 patients with frontal lobe epilepsy, including two patients with MRI-negative epilepsy. Interictal spikes were more abundant in MEG than those in the EEG recordings. Cluster analysis of MEG spikes followed by dipole localization was successful (n = 14) for twice as many patients as for EEG source analysis (n = 7). They concluded that the localizability of interictal MEG spikes in frontal lobe epilepsy was more robust than that of interictal EEG spikes. In a case study with atonic seizures due to the MRI-negative frontal lobe epilepsy,²⁴ MEG identified the original and propagated sources of MEG spike despite bilaterally synchronized EEG spikes. Akiyama and colleagues²⁵ similarly reported that MEG identified the leading spike dipole in a patient with cortical myoclonus and epileptic spasms secondary to a left frontal lesion. There are several reasons for the spatial and temporal resolution differences between EEG and MEG, and for explaining why MEG is able to distinguish the original spike sources among synchronously projecting and/or propagating discharges in neocortical epilepsy, especially in frontal lobe epilepsy: (1) a standard international 10–20 system EEG, with fewer sensors compared with the whole-head MEG sensor positions, may have insufficient spatial sensitivity to detect minor activation at the onset of spike; (2) the spatial distribution of MEG sensitivity over a minimum 3 cm² across the fissure²⁶ is estimated to be smaller than that of EEG, which requires > 6 cm²;²⁷ (3) temporal resolution of MEG is superior to conventional 19-channel scalp EEG in detecting the beginning of an interictal spike in the fissural cortex. After an interictal spike at a small area in the fissural cortex has spread to a wider area including the crown of the gyrus, the EEG may remain negative, or bilateral frontal spikes may sometimes appear, after the peak of the MEG spike.²⁸

Otsubo and colleagues²⁹ proposed the existence of a syndrome of “malignant Rolandic–Sylvian epilepsy (MRSE),” in contrast to “benign Rolandic epilepsy (BRE).” The MEG differentiated the characteristics of spike discharges between MRSE and BRE; MRSE features consisted of: (1) fronto-centro-temporal spikes on EEG; (2) refractory sensorimotor seizures; (3) MRI-negative; (4) neurocognitive deterioration; (5) MEG spike dipoles randomly oriented and located around the Rolandic–Sylvian fissures. Resective surgery based on MEG spike dipoles and icEEG can control the refractory epilepsy in MRSE. The MEG localized another type of nonbenign Rolandic–Sylvian spike dipoles in 18 patients with “atypical benign focal epilepsy (ABPE).” The ABPE showed bilateral MEG spike dipoles around the Rolandic–Sylvian fissures. These children may have continuous spikes and waves during sleep. Also, these children may need to switch from carbamazepine to ethosuximide to improve control of the multiple types of seizures which may include drop attacks, motor seizures, epileptic negative myoclonus, absences, and focal myoclonic seizures.

Paetau and colleagues³¹ observed that MEG is able to localize the epileptic discharges from the deep sulci, such as the Sylvian fissure. The MEG method records perpendicularly oriented magnetic fields generated by intracellular currents, whereas EEG preferentially records radially oriented electrical fields of extracellular currents. The palisading neurons in the wall of Rolandic–Sylvian interhemispheric fissures and deep sulci tangentially project the magnetic field to the brain surface and scalp to be detected by MEG.^{32, 33} Deep-seated neuronal activities in the deep sulci are not detected by scalp EEG. The extracellular currents radially projected from the brain surface obscure the deep-seated neuronal activities for EEG detection. The MEG signal, however, is reduced in inverse proportion to the square of the distance between the sensor and the current source.

The MEG showed showed regional differences in the sensitivity for detecting epileptic spikes: preferential regions with a high detection rate of electrocorticography-associated MEG spikes were fronto-orbital (100%), interhemispheric (89%), temporolateral (73%), frontal-superior (72%), and central (76%) regions, compared to the mesial temporal region (28%).³⁴ Occult peri-insular epilepsy in a small number of patients has been localized by MEG identification of epileptogenic regions, even when multiple other diagnostic methods had failed to do so.³⁵ There was either no interictal spikes or nonlocalizable ictal discharge on the scalp video-EEG in two of the three patients. The MEG localized the ictal discharges in the deep anatomical structure of the fusiform gyrus in an MRI-negative patient in whom scalp EEG failed to record ictal discharges.³⁶ Wang and colleagues³⁷ have also reported successful localization of seizures by MEG in a patient with subtle FCD in the frontal operuculum. They simultaneously recorded interictal and ictal discharges on MEG and stereotactic EEG, and these discharges were not detected on scalp video-EEG recordings.

The role of MEG in MRI-negative temporal lobe epilepsy

There was no report in the literature about the use of MEG specifically for MRI-negative temporal lobe epilepsy. For patients with mesial temporal lobe epilepsy (MTLE), MEG is not a method to exactly localize the spike source, because of the structural and functional configurations of the hippocampal formation³⁸ and other networks in temporal lobe epilepsy.^{39, 40} Therefore “dipole modeling” of MEG spikes has been applied to explain MTLE. Three dipole models of MEG spike have been described in MTLE: (1) anterior horizontal dipole, corresponding to mesial temporal discharges propagating to the anterior temporal pole; (2) anterior vertical dipole, corresponding to mesial temporal discharges propagating to the lateral temporal region; and (3) posterior vertical dipole, reflecting extensive temporal epileptic networks.^{39, 40}

Leijten and colleagues²² have compared whole-head MEG (151-channel) with high-resolution (64-channel) EEG for source localization in MTLE with unilateral hippocampal sclerosis. The sensitivity of MEG in detecting interictal spikes was only 32%, compared to 42% for EEG, and no patients showed MEG spikes only: MEG-localized sources were more superficial and EEG-localized sources were deeper. Magnetic signals attenuate according to the square of the distance from a source to a detector, and MEG prefers to detect open circuit neurons in the cortex of temporal lobes rather than closed circuit neurons in the amygdala and hippocampus. The MEG technique does not detect activities from the hippocampus itself, but detects discharges spread over the lateral and anterior part of neocortices including inferior, middle, and superior temporal gyri in most cases. In the rare situations of epileptic discharges appearing at parahippocampal or fusiform gyrus, MEG could detect the deep-seated cortical discharges.^{36, 38}

In contrast with the above findings, Kaiboriboon et al.⁴¹ have reported a high rate of MEG spike detection (85%) in 22 patients who underwent prospective MEG recording (whole-head 275-channel) and anterior temporal lobectomy with or without hippocampal atrophy on MRI. In 17 patients with unilateral hippocampal atrophy, MEG localized the spike sources ipsilateral to the site of surgery in ten patients (58.8%), including one with hippocampal atrophy contralateral to the side of a lobectomy. Two of four patients with MRI-negative temporal lobe epilepsy had localized spike sources ipsilateral to the side of surgery. The other two patients had bilateral or no spikes. The difference in sensor position or types, and the reduction of antiepileptic medications, may have contributed to the differences in the detection rates. Detection rates may also differ according to the types of MEG pick-up coils. Magnetometer coil was thought to be superior in recording spikes from deep regions including the mesial temporal lobe, and the planar gradiometer coil was thought to be superior in detecting superficial spikes, such as in the temporal neocortices.^{41, 42}

New analysis method

The single ECD concept is based on the assumption that there is only one compact current source in the brain. There are two mathematical steps: (1) forward problem, based on a design of the approximate model of the brain conductor, (2) inverse problem, in seeking the single ECD that can explain the measured magnetic fields with minimum error. When multiple neurons discharge synchronously in convoluted cortices and the discharge rapidly propagates to remote areas, there is a limitation for this ECD technique to delineate the epileptic discharges. Recently, various new techniques have been introduced to MEG analysis.

Synthetic aperture magnetometry (SAM) is a beamformer technique.⁴³ The SAM method is an adaptive spatial filter with the constraint that the variance of current fluctuation is minimized; SAM can estimate the temporal course of electrical neural activity at a particular site in the brain marked by 3D imaging voxels on MRI. Oishi et al.⁴⁴ applied this technique to MEG analysis to simulate intracerebral epileptiform discharges in the region of interest in ten pediatric patients, including four patients with MRI-negative epilepsy who underwent cortical resection based on icEEG. Oishi et al⁴⁴ compared interictal spikes on SAM with nonsimultaneously recorded icEEG. They demonstrated resemblance of the waveforms between virtual sensors on SAM and icEEG, at the same neocortices.

Sugiyama and colleagues⁴⁵ proposed an advanced version of SAM (SAM-g2) using kurtosis of reconstructed waveforms at the virtual sensors. Distribution of the automated high kurtosis voxels on MRI was consistent with the location of MEG spike dipoles from multiple/generalized/diffuse spikes surrounding the tubers in patients with tuberous sclerosis complex. The kurtosis reflects the combined effect of the amplitude and duration of interictal paroxysmus.⁴⁶ The SAM-g2 demonstrated the voxels with high kurtosis values as an epileptic region, possibly the ictal onset zone.

Mohamed and colleagues⁴⁷ applied the event-related beamformer method to estimate the spatial distribution of source power in the individual MEG spikes by comparing the MEG spike dipoles and the seizure onset zones on icEEG in 35 children with refractory neocortical epilepsy. The localization of interictal spikes corresponds to the centroid of MEG spike source cluster. The event-related beamformer localization was concordant with the seizure onset zone on icEEG at the gyral level in 69% of patients. In a subset of patients with unifocal event-related beamformer localization, beamformer results were concordant with the ictal onset zone on intracranial EEG in 22 of 23 patients.

Guggisberg and colleagues⁴⁸ applied short-time Fourier transformation to spike-locked waveforms at the virtual sensors. They reported the distribution of the virtual sensors with high beta and gamma activities is more suitably consistent with the epileptogenic zone than the distribution of ECDs.

Shriaishi and colleagues have expanded the indication of MEG not only in focal epileptiform discharges, but also in widespread epileptiform activities.^{49, 50} Dynamic statistical parametric maps (dSPM) are a spatial filter with the constraint that total weighted currents on the cortices are minimized. The dSPM reconstructed the propagation of spike or slow waves on the virtual sensors of the brain surface in two epilepsy patients.⁴⁹ They also applied the well-known short-time Fourier transformation process sensor by sensor to demonstrate the time evolution of the specific frequency bands during spike propagations in two patients with MRI- demonstrable lesion-related focal epilepsies.⁵⁰

Bouet and colleagues⁵¹ proposed a new procedure to delineate spiking volume. The volumetric imaging of epileptic spikes (VIES) with sorting of optimal high-frequency activities over 20 Hz is reconstructed with a recent beamforming approach, dynamic imaging of coherent neural sources, and statistical processing between spike periods and static baseline periods. They found a good agreement between VIES and intracranial stereotactic EEG (SEEG) in 21 patients with focal epilepsy.

Gradient magnetic-field topography (GMFT) is a method of inward projection of the signals of planar gradiometers vertically on to brain surfaces without solution of the forward or the inverse problem (Figure 6.3).⁵² Shizoru et al.⁵³ compared MEG, icEEG, and surgical outcomes and found that the distribution of the high-signal GMFT at the spike onset overlapped the ictal onset zone and interictal zone on icEEG better than focal/regional distributions of ECDs in neocortical epilepsy patients. The GMFT is less affected than ECD by signal-to-noise ratio when detecting with minimum error the area of spike onset.

Figure 6.3 Gradient magnetic-field topography (GMFT). Sequential GMFTs over the right hemisphere during 40 ms section (10 ms each) demonstrate the dynamic changes of the gradient magnetic field at one MEG spike. High gradients (red zones) represent the location of current sources of the spike. The red zone shifts inferiorly from the right frontal lobe towards the anterior temporal region during the single spike. Power bar at right represents gradient magnetic field (femtotesla/cm).

Fischer et al.⁵⁴ applied a novel technique designed to generate an ellipsoidal volume from the scattering of single MEG dipole localizations in 33 adult patients who underwent epilepsy surgery. The ellipsoidal volume analysis was compared with the resection volume generated from pre- and postoperative MRI images. A high coverage of the MEG ellipsoid volume by the resection volume and a low distance between the mass centers of both volumes correlated to a favorable seizure outcome.

Conclusion

In MRI-negative epilepsy patients whose scalp video-EEG shows electroclinical correlation that indicates focal-onset seizures, the colocalization of MEG spike dipoles with interictal EEG discharges is very valuable for supporting the eventual delineation of the epileptogenic zone by icEEG. Moreover, discrete epileptogenic networks could be suggested by the colocalization of FDG-PET and SISCOM findings with MEG findings. On the other hand, bilateral or scattered MEG spike dipoles in MRI-negative epilepsy require further investigational studies, including repeat MEG studies in some cases. Extensive or multiple epileptic networks are not infrequently encountered in MRI-negative patients. This situation is a major challenge for MEG localization of the potential epileptogenic zone, and also for the surgical resection of the zone.