Address correspondence and reprint requests to Dr. R.C. Burgess at Department of Neurology, Cleveland Clinic Foundation Desk S50, 9500 Euclid Avenue, Cleveland, OH 44195, U.S.A. E-mail: email@example.com
Summary: Purpose: Objectively to evaluate whether independent spike detection by human interpreters is clinically valid in magnetoencephalography (MEG) and to characterize detection differences between MEG and scalp electroencephalography (EEG).
Methods: We simultaneously recorded scalp EEG and MEG data from 43 patients with intractable focal epilepsy. Raw EEG and MEG waveforms were reviewed independently by two experienced epileptologists, one for EEG and one for MEG, blinded to the other modality and to the clinical information. The number and localization of spikes detected by EEG and/or MEG were compared in relation to clinical diagnosis based on postoperative seizure freedom.
Results: Interictal spikes were captured in both EEG and MEG in 31, in MEG alone in eight, in EEG alone in one, and in neither modality in three patients. The number of detections ranged widely with no statistical difference between modalities. A median of 25.7% of total spikes was detectable by both modalities. Spike localization was similarly consistent with the epilepsy diagnosis in 85.2% (EEG) and 78.1% (MEG) of the patients. Inaccurate localization occurred only in those cases with very few spikes detected, especially when the detections were in one modality alone.
Conclusions: Interictal epileptiform discharges are easily perceived in MEG. Independent spike identification in MEG can provide clinical results comparable, but not superior, to EEG. Many spikes were seen in only one modality or the other; therefore the use of both EEG and MEG may provide additional information.
Despite advances in imaging (both anatomic and functional), the most indispensable tools during diagnostic evaluation of epilepsy are the recording and interpretation of “brain waves.” These brain waves arise from summated dipolar electric currents generated in the cerebral cortex. Two fundamental methods exist to record brain waves—electroencephalography (EEG) and magnetoencephalography (MEG); each measures complementary manifestations of the same phenomenon.
The EEG is traditionally “read” by experienced interpreters, by using an appreciation of waveform appearance to discriminate epileptiform activity from background and other nonepileptiform activities based on waveform morphology, distinctness, and experiential knowledge (1,2). No independent confirmation of true positives or true negatives yet exists, and human interpretation is still the gold standard for the identification of spikes (2). The spatial correlation of the irritative zone (area from where interictal spikes arise) with the epileptogenic zone can help to define whether an EEG transient is epileptogenic. A gold standard for the correct localization of a spike also is difficult to find in vivo, but the elimination of seizures after resection of a cortical region strongly supports the notion that the tissue removed included the spike focus. The same limitations apply to the interpretation of MEG. Therefore as with EEG, it is reasonable to identify MEG spikes by independent analysis of the MEG alone, and then to correlate the findings with relevant clinical information. Surprisingly, very few studies have addressed this issue critically (3–5).
MEG has been increasingly used for localization of epileptic activity (6–17). Most of those articles show only “abstractions of the activity” as maps or data points superimposed on the magnetic resonance imaging (18). The original data (waveforms) and the method of spike detection are sometimes only perfunctorily described. Without presentation of the raw data and all of the well-known nuances that accompany visual interpretation, the neurophysiology community at large has been having difficulty assessing the usefulness or reliability of MEG relative to EEG (18). MEG is ordinarily used more precisely to estimate the source of spikes that have already been identified in the simultaneously recorded EEG. Some reports, however, describe the presence of spikes seen on MEG that were not captured with scalp EEG (6,15,19,20). The fact that several authors have reported some spikes seen in one modality but not in the other begs the question of whether MEG is superior for the detection of some epileptic transients or whether it simply picks up more nonepileptic transients. It is, therefore, important to clarify whether either modality is superior in terms of detection of epileptiform activity, what characterizes the different abnormalities detected by each modality, and whether these findings provide additional, clinically relevant information. To compare their detection capabilities, it also is necessary to know whether independent review of MEG provides clinically valid or erroneous results.
Although the retrospective study of consecutive intractable focal epilepsy patients cannot provide comprehensive answers to all of these questions, our primary purpose is to ascertain whether independent MEG analysis (without knowledge of the clinical information) is valid clinically. The second purpose is to investigate qualitative difference in the detection of spikes by EEG and MEG (if any).
The retrospective data included 43 patients who were evaluated at Kohnan Hospital, Japan, and fulfilled the following criteria: (a) underwent long-term video-EEG monitoring for presurgical evaluation of epilepsy, (b) were tested between September 1998 and December 2000 (n = 76), (c) underwent at least one session with simultaneous EEG and MEG recordings (n = 74 of 76), (d) proceeded to epilepsy surgery, and (e) had ≥1 year of follow-up. One patient was excluded from this study because of data loss.
The patients consisted of 25 male and 18 female subjects, ranging in age from 5 to 47 years (mean age, 26.0 years). Thirty-six patients had temporal lobe epilepsy (TLE), three patients had frontal lobe epilepsy (FLE), two patients had frontotemporal lobe epilepsy (FTLE), and two patients had parietal lobe epilepsy (PLE). Pathologic diagnosis included hippocampal sclerosis in 26 (Table 1). One year after surgery, 35 patients were seizure free (Engel's class I) (21), five patients had rare seizures (class II), and three patients had a worthwhile improvement (class III).
Table 1. Patient characteristics
*Patients with sphenoidal electrodes.
†Only nonconsistent (overlapping or inconsistent) results were noted by N.
Scalp EEG and MEG were measured simultaneously by using the same method described in previous studies (9,22). MEG was recorded from planar gradiometers either sitting with a Neuromag 122 system (n = 37) with 122 sensors, or supine with a VectorView MEG system (n = 6, patients between September 1999 and January 2000; Neuromag Ltd., Helsinki, Finland) with 204 sensors. EEG was recorded from 23 electrodes applied to the scalp according to the international 10–20 electrode system with sphenoidal electrodes (SPs) (n = 14), anterior temporal electrodes (ATs) (n = 27), or no special electrodes (n = 2). Spontaneous activity was recorded for an average of 21.2 min (range, 10.3–37.3 min), divided into epochs of 3.83 ± 1.45 min. The sampling rate was 400 samples/s/channel.
Two experienced investigators (authors M.I. and E.P.) independently reviewed the raw EEG and MEG signals (i.e., time vs. amplitude waveforms). Both were trained in EEG and epilepsy for a minimum of 2 years at the same institution (Cleveland Clinic Foundation) and had substantial experience in distinguishing epileptiform discharges from other, nonepileptic, sharply contoured transients. Investigator E.P. was assigned to review all of the EEGs. Investigator M.I., who had prior experience in both MEG and EEG, was assigned to read all of the MEGs. Although separated by 1 year from the present study, M.I.'s MEG training included significant exposure to physiologic MEG transients such as asymmetric sleep vertex waves and prominent temporal theta rhythms. Both investigators were blinded to the patient name and clinical information. The recordings were made anonymous by assigning a random ID number to each patient. Then the order of the data presentation was randomized independently for EEG and MEG. The investigators also were blinded to the results of the other modality by eliminating the EEG channels and MEG channels from the display. The reviewers were able to select different montages and to move backward and forward in time within the record to confirm their findings. The same display bandpass filter setting was used for reviewing both EEG and MEG (0.5–70 Hz).
For review of the EEG waveforms, the standard institutional (Cleveland Clinic) montage set was used (including longitudinal bipolar with/without additional AT or SP electrodes, transverse bipolar, ipsilateral ear referential, and Cz referential montages). Epileptic spikes or sharp waves were manually detected by using conventional visual identification, as is done during routine EEG interpretation (23). The time point of the peak was noted for every epileptic spike or sharp wave.
All EEG findings, including locations of the spikes, also were summarized into an EEG report form according to the institutional EEG classification (23). All abnormal findings including epileptic transients and other slow-wave activities were classified into regional or generalized activity along with anatomic location or laterality of the maximum. This summary result was used in later comparison as the “manual EEG interpretation.” This type of summary classification is routinely used in our institution during the process of clinical diagnosis.
During review of the MEG waveforms, a raw spatial montage set was used, in which all sensor pairs were divided into eight groups: frontal/temporal/vertex/occipital region in the left and right side each. To review all channels, it was necessary to switch the montage display 8 times. Epileptic spikes or sharp waves were manually detected by applying the same general principles recommended by the International Federation of Clinical Neurophysiology (IFCN) and used in standard EEG interpretation: sharp transient clearly different from background activity with an “epileptiform” morphology and a logical spatial distribution (24). The time point was noted for every epileptic transient.
After all spikes had been independently identified, every spike event was classified as one of the following: (a) unique EEG spike (EEG spike not detected in MEG), (b) unique MEG spike (MEG spikes not detected in EEG), or (c) common EMEG spikes (same spike detected in both modalities) (Fig. 1). Consequently, total spikes are represented by the sum of unique EEG spikes, unique MEG spikes, and common EMEG spikes. Total EEG spikes consisted of unique EEG spikes plus common EMEG spikes. Total MEG spikes consisted of unique MEG spikes plus common EMEG spikes. Spikes were objectively characterized as common when the time points of the peaks marked in EEG and MEG were within 100 ms of each other.
Computerized source estimation was performed independently for EEG and MEG, resulting in four different categories of spike localization: (a) EEG source estimation for the common EMEG spikes, (b) MEG source estimation for the common EMEG spikes, (c) EEG source estimation for unique EEG spikes, and (d) MEG source estimation for unique MEG spikes (Fig. 2).
Source estimation was based on five to eight representative spikes chosen from each group: common EMEG spikes, unique EEG spikes, and unique MEG spikes. Spikes having sufficient signal-to-noise ratio were selected by visual inspection. A 120- to 160-ms time epoch around each spike peak was segmented for the source analysis.
Spike localization was estimated with minimum-norm current image by using BESA2000 software, version 4.2.12 (MEGIS software GmbH, Munich, Germany). The current distribution of a fixed dipole set within a modeled brain was approximated to the measured data with the least square minimum norm method (25,26). A predetermined set of two layer dipole grids, which represented superficial and deep cortical areas, was assigned in the standard realistic head model. EEG calculation was performed in the standardized realistic volume conductor model by using a predetermined standard electrode location. For MEG calculation, the spherical volume conductor was derived from the patient's brain magnetic resonance imaging (MRI). The center of the head was coregistered to the measured data by using the 3-dimensional digitization technique that was done at the time of measurement. The current intensity distribution of the dipoles was then displayed by using a percentage scale on a surface image of the standard brain (Fig. 2).
We decided to use minimum norm estimation, instead of the conventional dipole modeling, because this method does not require an initial guess by the investigator. By avoiding any interpreter's bias, the method yields objective results for both EEG and MEG. Unlike the point-like answer generated by the dipole model, however, this method provides only a plausible current distribution extending over the cortical surface.
Review of the source estimation and comparison to epilepsy diagnosis
The results of the source estimation were reviewed by three investigators (M.I., E.P., R.C.B.) with no exposure to clinical information. Because each spike may have different characteristics, the localization was discussed after presentation of all spike samples up to five. The source localization was determined for each spike category based on the lobe of maximal current density. If not confined to a single lobe, localization was denoted by naming two or more lobes together. If the activity was identified on one hemisphere, but not confined to any lobes, it was classified as hemispheric. If the source estimation did not indicate any hint of regional activity, it was classified as nonlocalizable.
Epilepsy diagnosis was established independent of this study based on a convergence of multiple diagnostic evaluations. All the patients underwent resective surgery according to the diagnosis. Association of the source localization to epilepsy diagnosis was assessed on a lobar basis. It was defined as “consistent” if the distribution shared the same lobe and the difference did not exceed one lobe, as “nonconsistent” if the difference was greater than one lobe, if the distribution did not share the same lobe, or if the source estimation was nonlocalizable.
Wilcoxon's signed-rank test was used to evaluate the difference between two nonnormally distributed paired continuous data.
Spike detection and intermodality comparison
Interictal spikes were detected in both EEG and MEG in 31, in MEG alone in eight, in EEG alone in one, and in neither modality in three of a total of 43 patients. One patient (Pt 17, Rt FLE), whose spikes were detected by both EEG and MEG, was excluded from further analysis because identification of single epileptiform discharges was apparently blurred by a morphology that was characterized by beta-range polyspike bursts and spike–wave complexes lasting >3 s. All the other cases had isolated epileptiform transients. Figure 2 shows typical examples of common and unique spikes, and the current density images. Figure 3 shows the distribution of EEG and MEG spike populations in the 39 patients who had interictal spikes. The median number (range) of detections was 76.5 (0–1,057) total spikes, 19.0 (0–875) total EEG spikes, and 35.0 (0–600) total MEG spikes. No statistical difference was found between EEG and MEG (n = 42, Wilcoxon's signed-rank test; p = 0.81). The use of sphenoidal electrodes was not associated with a larger number of spikes detected in the EEG.
EEG and MEG spike proportions ranged widely (Fig. 3). The 30 patients who had interictal spikes in both EEG and MEG can be roughly divided into two groups: those with a much larger number (>70% of total spikes) of EEG spikes and those with larger number of MEG spikes, although continuity was seen between the two. Common EMEG spikes accounted for a relatively small proportion of the total spikes (median, 25.7%, with a range of 3.6–83.3%; Table 2). Unique EEG and MEG spikes represented 16.4% and 34.5% (median) of total spikes. This difference was not statistically significant (n = 30, Wilcoxon's signed-rank test; p = 0.62 for the number, p = 0.47 for the percentage population, respectively). Common EMEG spikes accounted for median 51.1% (3.8–100.0) of total MEG spikes and for median 60.0% (7.9–100.0) of total EEG spikes, respectively. These ratios are a measure of the relative detection sensitivity of each modality. Both values ranged widely with no statistical difference.
Table 2. Spike population in 30 patients who had both EEG and MEG spikes
Number of spikes
Percentage to total spikes (%)
6 – 1057
Total EEG spike
1 – 875
5.3 – 100.0
Total MEG spike
5 – 600
7.9 – 100.0
Common EMEG spike
1 – 425
3.6 – 83.3
Unique EEG spike
0 – 489
0.0 – 92.1
Unique MEG spike
0 – 363
0.0 – 94.7
Spike lateralization/localization and epilepsy diagnosis
Clinical characteristics of the patients are summarized in Table 1. We assessed the relation of spike localization to epilepsy diagnosis only in the 35 patients in whom the preoperative diagnosis was confirmed by seizure freedom after surgery. Preoperative diagnosis included TLE in 32, FLE in two, and FTLE in one patient. Among them, 20 had mesial TLE with hippocampal sclerosis. Interictal spikes were identified in both EEG and MEG in 26, in MEG alone in six, in EEG alone in one, and in neither EEG nor MEG in two patients. When the analysis was limited to the patients with TLE or mesial TLE, no changes were found in the general results described.
Hemisphere of spikes localized
When EEG and MEG independently identified a spike occurring at the same time (i.e., common EMEG spike), the hemisphere of the activity was never incongruent between EEG and MEG. In 23 patients, EEG and MEG identified only unilateral spikes, all in the epileptogenic hemisphere. In the seven patients who had bilateral spikes, neither modality had a clear superiority in the detection of spikes from the epileptogenic hemisphere. In three patients (patients 5, 27, and 38), who had a very small number (fewer than five) of spikes which occurred only in one modality (i.e., EEG or MEG alone), the spikes were identified in the hemisphere contralateral to the diagnosis.
Localization of spikes
EEG spike localization was consistent with the epilepsy diagnosis in 23 (85.2%) of 27 patients (the denominator is the number of patients who had any spikes). MEG spike localization was consistent with the epilepsy diagnosis in 25 (78.1%) of 32 patients. Inconsistent localization (i.e., nonlocalizable or no shared lobes), was seen in nine patients (numbers 5, 6, 8, 27, 28, 31, 38, 40, and 42) in whom the number of spikes detected was fewer than 19. The occurrence of inconsistent localization had no clear association with pathologic findings or distribution of mesial and lateral temporal lesions.
The manual EEG interpretation was consistent with the epilepsy diagnosis in all except six (77.8%) patients. EEG spike location was read as frontal for TLE in three (patients 18, 31, and 36), as parietal for FLE in one (patient 32), as frontotemporoparietal for TLE in one (patient 21), and as right frontotemporal for left TLE in one patient (patient 38). Interestingly, computerized source estimation by the current density image was consistent with the diagnosis in four of those six patients, including the patient with FLE.
Clinical value of unique spikes
The localization accuracy of the unique spikes was generally the same as or inferior to the common spikes. The unique EEG spike was superior in one patients (number 2), and the unique MEG spike was superior in one patient (number 37). The latter patient had only one common EMEG spike against 25 unique MEG spikes.
When a small number of spikes was found only in one modality, the relation of spike localization to epilepsy diagnosis was less certain. In the six patients with MEG spikes alone, the localization was consistent with epilepsy diagnosis in only two patients. In the one patient with EEG spikes alone, the spikes were localized in the hemisphere contralateral to the diagnosis.
Patients who were not seizure free
Seven patients, who had residual seizures at 1-year follow-up, were excluded from the analysis. The seizure outcome included Engel's grade II in four patients with TLE and one patient with PLE, and grade III in one patient with PLE and one patient with FTLE. Interictal spikes were seen in both EEG and MEG in four patients, in MEG alone in two, and in neither modality in one. In the two patients with PLE, 37 and 117 spikes were captured only by MEG and localized concordant to the diagnosis. In these two, manual EEG interpretation was reported as normal in one and as bilateral frontotemporal slowing in one.
Clinical validity of independent MEG interpretation
This is the first study of independent MEG spike detection that has been performed blinded to the clinical and other neurophysiologic information. This study shows that visual identification of epileptic spikes from MEG is similar to EEG (i.e., independent detection of MEG spikes is as reliable as traditional independent detection of EEG spikes). In the patients with a seizure-free postsurgical outcome, no significant difference was found in clinical correlation between EEG and MEG spikes at lobar-level localization. Only in those few patients with a relatively small number of spikes, especially when the detections were in one modality alone, spike localization tended to be misleading. More prolonged recording might have produced a larger and more informative sample of spikes.
These results are specifically applicable to our patient population (consisting mainly of TLE patients) and perhaps cannot be easily extrapolated to the general population of patients being evaluated for epilepsy surgery. We eliminated about 30 patients who underwent MEG, but who were not operated on. Because the surgical decision, taken independent of our study, took account of both EEG and MEG findings, our study could be biased because (a) patients who had incongruent results between modalities may not have been offered surgery, (b) localization becomes self-fulfilling because the surgeon is likely to take out all of the areas implicated by either or both modalities. The inclusion of only the operated patients, although necessary to measure localization accuracy, probably favors those patients where EEG and MEG were more congruent.
The yield of MEG in terms of the number of spikes is the same as EEG. In this study, both EEG and MEG were simultaneously recorded for a relatively short period, ∼20 min. EEG and MEG are generally used in different ways, and it is difficult to compare our results with the yield of typical EEG. However, the longer recording time usually afforded by long-term monitoring may favor EEG's chance to detect epileptic spikes. It is unlikely that patients who have rare or no epileptiform activity during long-term EEG will manifest spikes on MEG, and some reviews have suggested that MEG should generally be used for patients with frequent spikes on scalp EEG (27). However, it is worthwhile to point out that eight of 40 patients had MEG spikes only. Particularly interesting is the observation that two patients had >30 spikes that were recorded only in the MEG and were localized in the parietal region consistent with the diagnosis. Conversely, seizure outcome was poor in both of these cases suggesting that this lack of convergence (i.e., no common EEG-MEG spikes) decreases the reliability of the information. The specificity of MEG is still unknown because we did not include control subjects in this study. MEG captured spikes not relevant to the clinical diagnosis in four of eight patients who had MEG spikes alone, whereas MEG agreed with EEG in terms of absence of spikes in three patients. It also is likely that interinterpreter inconsistency is more frequent for MEG than for EEG because only limited experience is found with visual MEG spike identification (5).
At a lobar level of comparison, the localization ability of MEG is the same as EEG. However, this study did not evaluate the MEG localization ability at the sublobar level. Although refined localization ability is often thought to be one of MEG's main advantages, we intentionally avoided using dipole source modeling (the most common method in practice) to eliminate interpreter bias. Dipole modeling always provides point localization and may be useful to elucidate small localization differences. However, the dipole model tends to be erroneous for the extended generators frequently seen in epileptic spikes.
This study demonstrates that the empirical knowledge used in reading EEG also is acceptable for use in MEG. However, several caveats must be taken into consideration. Some rules of thumb, such as “phase reversal” technique, are meaningless in MEG but useful in EEG to depict the distribution of activity manually. Magnetic field distribution is usually drawn after computerized calculation from hundreds of sensors. Planar-type gradiometers provide maximal peak amplitude close to the source location when a single dipolar generator is assumed (28) but this may not be the case in the other sensor configurations. MEG has a different vulnerability to external noise (29) and has sensitivity to some uncommon background brain noise (30,31). Therefore some additional knowledge may be required for reliable MEG interpretation.
Our study subjects were relatively homogeneous in regard to the diagnosis of TLE. The characteristics of MEG/EEG spike detection may be different in other types of epilepsy (12,14). The number of patients with extratemporal lobe epilepsy in our study is insufficient to address this question.
Reasons for differences in the number of spike detections
The fact that a number of unique spikes were detected in only one modality suggests that EEG and MEG may be complementary. Because we performed only one-time review of EEG and MEG, each by only one interpreter, it is difficult to define the proportion of variability caused by intermodality difference versus interinterpreter difference. Interinterpreter variability in detection may partly explain our low fraction of common EMEG spikes. However, previous studies also reported significant differences in spike detection between modalities. In a study of 19 MTLE patients considered good candidates for epilepsy surgery who had simultaneous MEG/EEG recordings of a duration similar to ours, Zijlmans et al. (5) found unequivocal spikes (defined as significantly high agreement among three observers) in EEG alone in two patients, in MEG alone in two, and in both EEG and MEG in only one. Lin et al. (32) classified interictal spikes during simultaneous EEG and MEG recording similar to our study. They reported that a large proportion (68.4%) of spikes was visible in both EEG and MEG, but both EEG and MEG were reviewed simultaneously by one interpreter in their study. Both Lin et al. and Zijlmans et al. noted that simultaneous observation of EEG and MEG influences the spike identification in each modality. Use of information from both modalities may increase the sensitivity and specificity to spikes (18).
It is possible that the orientation of the spike generator played a major role in the different detectabilities. Theoretically, MEG exclusively detects tangential vector components of generators, and the EEG is relatively more sensitive to radial components (28,33–35). In actual practice, however, we encounter a broad range of orientations, with the real generator extending over the cortical surface, including both gyral wall and lateral surface (7). So MEG not only detects tangential “generators,” but also tends to see the tangential part of these dipole sheets along the cortical surface. The cortical areas that contribute to surface EEG and MEG activities largely overlap each other (36). As the study by Malmivuo et al. (35) shows, the size and distribution of the volume that a planar gradiometer is sensitive to is comparable to that of scalp EEG recorded from a pair of neighboring electrodes in the standard 10–20 placement. Just as in EEG, the primary determinant of the MEG sensitivity to brain activity is the distance from the sensor.
In our blinded study, each reviewer was required to look for transients and to make a “spike” or “nonspike” decision about every one. In truth, however, the reviewer is actually going through a less binary process of deciding whether a given half-wave exceeds his threshold for abnormality (which is dependent on the local signal-to-noise ratio as well as the severity of the underlying disorder). In retrospect, not all of the unique spikes were completely invisible in the other modality, but instead were just less distinct in one modality than the other, and hence were not marked by the investigators. As we do when reviewing the results of most other diagnostic tests, it is important to observe EEG and MEG separately and to identify carefully the abnormalities in each modality.
Acknowledgment: The author (M.I.) is partly supported by Uehara Memorial Foundation. We thank for technical assistance and data transfer to Prof. Keisaku Hatanaka, Department of Applied Physics, Okayama University of Science, and Akitake Kanno, Kohnan Hospital, Japan. We also are grateful to Drs. Michael Scherg and Karsten Hoechstetter, MEGIS software GmbH, for their kind advice on software use.