Identification of the Epileptogenic Tuber in Patients with Tuberous Sclerosis: A Comparison of High-resolution EEG and MEG


Address correspondence and reprint requests to Dr. F.E. Jansen at Department of Neurology, C03236, University Medical Center, PO Box 85500, 3508 GA Utrecht, The Netherlands. E-mail:


Summary: Purpose: We compared epileptiform activity recorded with EEG and magnetoencephalography (MEG) in 19 patients with tuberous sclerosis complex (TSC) and epilepsy.

Methods: High-resolution (HR) EEG, HR-MEG, and 1.5-T MRI scans were performed. Epileptiform spikes were identified in EEG and MEG recordings offline by three observers. Spikes for which the interobserver agreement (spike consensus) was >0.40 were used for source localization with CURRYV 3.0 software. MUSIC analysis was performed. The distance between the source determined from EEG and MEG recordings and the border of the closest tuber was calculated and compared.

Results: Consensus spikes (kappa >0.4) were identified in 12 patients in the EEG recording and in 14 patients in the MEG recording. MEG sources were closer to tubers in all but one patient. Three patients underwent epilepsy surgery, two of whom are seizure free after complete resection of the tuber.

Conclusions: In patients with TSC, epileptogenic sources identified on MEG are closer to the presumed epileptogenic tuber than are similar sources identified on EEG. Moreover, spike consensus is greater with MEG. Clear identification of the epileptogenic zone may offer opportunities for surgery in patients with TSC with intractable epilepsy.

Tuberous sclerosis complex (TSC) is a neurocutaneous syndrome, involving multiple organs. The characteristic hamartomas are most commonly found in the skin, retina, heart, kidney, and brain. TSC is an autosomal dominant disorder with linkage to chromosome 9q34 (TSC1) (1) and chromosome 16p13 (TSC2) (2). Hamartin and tuberin, the protein products of TSC1 and TSC2, respectively, are tumor-suppressor genes.

In the CNS, the disordered proliferation, migration, and differentiation of neurons as a result of TSC give rise to noduli, subependymal giant-cell astrocytomas, and cortical tubers (3). Cortical tubers are associated with neurologic symptoms, such as epilepsy, mental retardation, and focal neurologic deficit. Although the phenotypic expression of TSC is extremely variable, seizures are common, occurring in 80–90% of cases, and are often the presenting symptom. Furthermore, they are often intractable (50%). Surgery should be considered in patients with TSC and drug-resistant epilepsy, but it may be difficult to identify the epileptogenic tuber if several tubers are distributed throughout the cerebral cortex. To date, the outcome of surgery has been variable in children with TSC (4–10).

Although structural and functional imaging techniques [MRI, functional MRI (fMRI), single-photon emission computed tomography (SPECT), positron emission tomography (PET)] are being increasingly used in patient workup, epileptiform activity can be recorded only with neurophysiologic techniques. With standard EEG, it is often impossible to delineate the irritative zone and establish its relation to the tuber(s) with sufficient precision, and for this reason, high-resolution (HR) EEG and HR-magnetoencephalography (MEG) are used to increase the accuracy of functional localization (11). Source localization with EEG requires modeling of the skin, skull, and brain, which in turn requires knowledge of their shape and their conductivity. Conductivity values in particular can only be approximated. Because MEG records the magnetic field around the head, and magnetic fields are not attenuated by volume conductors, MEG is more accurate than EEG for source localization. Results from the literature based on realistic phantom models, using the same methods (MUSIC) as described in this article and for comparable numbers of measurement channels, indicate that localization errors for true dipolar sources on average are 7–8 mm for EEG and 3 mm for MEG (12). However, because a radial dipole does not generate a magnetic field outside the head, MEG selectively detects tangential sources (e.g., sources on a sulcus). Moreover, measurements are affected by movement of the head, which means that patient cooperation is essential. The shortcomings of EEG and MEG can largely be overcome by combining the two techniques, by recording MEG and EEG simultaneously. The sources computed on the basis of HR-EEG and HR-MEG findings can be visualized by plotting the equivalent current dipole representation on the MRI of the patient's brain with volume reconstruction. This technique, termed magnetic source imaging (MSI), has been successfully applied to reduce the need for invasive monitoring in candidates for surgery (11,13,14). Only limited experience has been gained in the use of MEG in patients with TSC and epilepsy (11,15,16).

In this study, we determined the correspondence between epileptiform activity recorded with EEG and MEG in patients with TSC and epilepsy and to what extent the sources differed after mapping of the relevant dipoles on MRI images.



The study included 19 patients with TSC who fulfilled the revised TSC diagnostic criteria and who were admitted to the outpatient clinic of the Departments of Pediatric Neurology and Neurology, University Medical Center, Utrecht, The Netherlands, in the period from 1998 to 2003. All but one patient had active epilepsy, and all had relatively frequent interictal spikes on HR-EEG. A seizure history was taken, and further information was extracted from the medical files.



Cerebral MRI was performed on all patients by using a 1.5-Tesla magnet (Gyrocsan NT, Powertrak 6000; Philips Medical Systems, Best, The Netherlands). Fluid-attenuated inversion recovery (FLAIR) images were generated with a slice thickness of 1.5 mm, no gap. Optimal scan parameters were determined in a pilot study (IR, 2,600 ms; TE, 125 ms; TR, 11,000 ms; FOV, 256 × 256; RFOV, 80); the contrast-to-noise ratio with a slice thickness of 1.5 mm is sufficient to detect small tubers. On the FLAIR images, tubers were semiautomatically segmented by using imageXplorer []. The interface requires manual identification of each tuber by a single click inside the tuber. The local maximum (highest intensity value) inside the tuber is then used as starting point for a region growing procedure. The lower threshold of the region growing range (that defines the border of each tuber) can be determined automatically based on statistical analysis of previously segmented tubers in the same data set (mean value, standard deviation). With this learning system, the majority of the bright tubers have been segmented with a single mouse click. The tuber load was expressed as number and volume of tubers. For the construction of the volume conductor models, T1-weighted images, with a resolution of 0.89 × 0.89 × 1.5 mm, were recorded. FLAIR images were matched to these according to a previous published image-registration protocol (17).

EEG and MEG spike acquisition

An EEG (85 channels; BioSemi Mark-6, Brainstar system 4.0) was recorded in 14 patients at a sampling rate of 1,024 Hz, by using a cap containing Sn electrodes (ElectroCap Inc.). Furthermore, when MEG was performed, a 64- or 32-channel EEG was recorded at the same time in all patients. For the purpose of this study, the highest quality EEG recordings with the lowest number of bad leads and the highest signal-to-noise ratio were analyzed. In all cases, this was the 85-channel recording. Only in those cases in which such a recording was not available, or in which no spikes were detected, was the EEG recorded simultaneously with MEG (also) analyzed. Thus in 14 patients, the 85-channel EEG recording was analyzed, and in five patients, the 64- or 32-channel EEG recording, performed simultaneous to the MEG recording, was analyzed. The electrodes of the HR-EEGs were positioned according to the 10% system accepted by the American EEG society. Electrode positions were registered by using a magnetic tracking device (Polhemus, Colchester, VT, U.S.A.). In addition, three anatomic marker points (e.g., nasal and preauricular) and head shape were measured, which allowed matching with MRI and MEG markers. Spontaneous activity was recorded during a 40-min session.

MEG, using 151 axial gradiometers arranged as a helmet (Omega 151; CTF system Inc., Port Coquitlam, BC, Canada), and simultaneous 32- or 64-channel EEG were recorded at a sampling rate of 625 Hz, inside a magnetically shielded chamber (vacuumschmelze GmbH, Hanau, Germany). Head position with respect to the helmet and, after each recording session, electrode positions and head shape were recorded, using four magnetic localizing coils (18). The signals were (software) bandpass filtered between 0.7 and 70 Hz. A recording session typically took 1.5 to 2 h. Aside from voluntary sleep deprivation, no activation procedures were used. None of the patients needed sedation.

Spike detection and averaging

Interictal epileptiform spikes were identified offline in EEG and MEG recordings. Epileptiform activity was analyzed for selected 10-min epochs; two or three 10-min MEG epochs were analyzed. Epileptiform activity was defined as sharp activity different from background activity and lasting <0.1 ms. Sharp activity registered during the QRS complex of the electrocardiography (ECG) was neglected. Three observers independently analyzed the selected EEG and MEG epochs for the 19 patients to obtain consensus spikes (spikes selected by two or more observers) (19,20). The observers were not given information about seizure semiology, MRI findings, and spike selection by the other observers. Kappa statistics were calculated between all three combinations of the three observers for both EEG and MEG recordings. Recordings were rejected when the interobserver kappa values were <0.4 for two or more combinations of the three observers. For EEG and MEG recordings with kappa values ≥0.4, cluster analysis of the marked consensus spikes was performed (21). Spikes for each cluster were then averaged. In the case of multiple clusters, only the one with the highest number of spikes, resulting in the highest signal-to-noise ratio, was considered for source imaging.

Source imaging

Spherical volume-conductor models are inadequate for accurate source localization, especially in the case of EEG. Therefore based on the 3DT1 MRI, an accurate volume-conductor model including brain tissue (MEG only) and skull and skin (EEG) was constructed by using CURRY 3.0 software. The conductivity of the skin, bone, and brain compartments was set at 0.33, 0.0168, and 0.33 mS/m, respectively. Measured EEG electrode positions and position of the MEG helmet with respect to the volume-conductor model were computed by using the co-registered marker positions and head shape. Averaged spike data (for an ∼30- to 40-ms period from spike onset to spike maximum) were imported into CURRY. A MUSIC scan analysis for rotating dipoles was then performed at a resolution of 1.5 mm. The rank applied was that suggested by an SVD analysis of the spike selection, assuming a noise level as indicated by the data before spike onset. The distances between the dipole maximum identified by the MUSIC analysis of both EEG and MEG data and the border of the closest tuber were calculated and compared.


Patient characteristics

Table 1 summarizes the clinical data of the 19 patients (11 female, eight male patients). The diagnosis of TSC was confirmed by mutation analysis in 11 patients (six had the TSC1 mutation, and five, the TSC2 mutation). Median age at seizure onset was 1 year (range, 1 month to 37 years). Median age at the time of the study was 27 years (range, 6 to 54 years). Thirteen patients had complex partial seizures; three, tonic seizures; and six, secondarily generalized tonic–clonic seizures. One patient had been seizure free for >20 years but showed relatively frequent interictal epileptiform discharges. Two patients had daily seizures; 11 patients several seizures per week; and five patients had seizures at least once per month. Four patients had more than one seizure type at the time of investigation. In the patients with only partial seizures, seizure semiology had not changed for many years. Fourteen patients were examined neuropsychologically: full-scale IQ ranged from <48 to 119; median, 78. In 20 patients, FLAIR images allowed the segmentation of tubers. The median number of tubers was 18.5; range, 5 to 45.

Table 1. Clinical characteristics of 19 tuberous sclerosis complex patients
PtAge at exam (yr)Age at Sz onset (yr)Sz typeSz frequencyIQMutationNo. tubers/vol in cc
  1. sz, seizure; IQ, intelligence quotient; CPS, complex partial seizure; sGTCS, secondarily generalized tonic–clonic seizure; TS, tonic seizure; SPS, simple partial seizure; 0, seizure free; 1, monthly; 2, weekly; 3, daily; vol, volume.

1 62  CPS283TSC121/4.9
2530.1sGTCS2<48  24/0.7
3320.9TS, sGTCS1 TSC219/1.0
4190.7TS, sGTCS373TSC112/2.9
52312   CPS1 TSC24/11.9
7541  sGTCS057 16/0.2
10210.8sGTCS2 19/6.0
112827   CPS2 25/19.8 
122110   CPS297TSC113/0.2
13175  CPS2119 TSC112/1.8
152714   CPS, TS2  8/0.1
17300.4CPS267 18/0.6
18279  CPS, SPS3117 TSC129/15.8 
196037   CPS278 11/0.3

Spike detection

In all 19 patients, only interictal spikes were obtained (Table 2). EEG recording revealed a single localization of epileptiform activity (spikes) in nine individuals. The interobserver analysis of all EEG recordings showed kappa values >0.40 for at least two combinations of observers in 12 patients. The number of consensus spikes in the EEG recording ranged from 5 to 168 (median, 38). In those recordings for which poor interobserver agreement was found, we compared the 85-channel EEG with the EEG recorded at the same time as the MEG was recorded. Spike detection and agreement were not better for the simultaneous EEG recording. In two patients, EEG spikes were absent, although spikes were detected in the MEG recording. MEG recordings showed unifocal epileptiform activity in 12 patients. The kappa values exceeded 0.40 for the MEG recordings of 14 patients. In the MEG recordings, the median number of consensus spikes was 29 (range, 8–163). Kappa values were low for the recordings, with few consensus spikes.

Table 2. Spike detection in 19 patients with tuberous sclerosis complex
KappaSpikesFociDistance (mm)KappaSpikesFociDistance
  1. U, Unifocal; M, multifocal; in brackets is the distance between the source and the tuber closest to the other modality.

1>0.4140 U 5.5>0.415U7.4
2>0.4126 M28.4>0.484U12.1
3>0.497M15.4>0.4122 M6.7
4>0.492U21 (26.3)   >0.470U20.1 (46.2)
6>0.470U22.5 (34.4)    >0.434U10.4 (32.3)
9>0.4127 M19.2<0.412 
10>0.432M33.7<0.4 8 
11<0.436 >0.4163 U11.7
12<0.4125  >0.480U13.1
13<0.4 8 >0.429U10.8
14<0.414 >0.422U7.6
15>0.4168  >0.4123  
16>0.429 >0.429 
17<0.416 <0.411 
18<0.414 <0.416 
19<0.4 5 <0.415 
Average 24.6 (26.4) 13.8 (21.3) 

Distance to tubers

The EEG and MEG epileptic sources were plotted on the brain images for eight patients, EEG sources only for two patients (patients 9 and 10), and MEG sources only for four patients (patients 11–14). Source modeling could not reliably be performed in two patients even though interobserver agreement existed about the epileptic source for both EEG and MEG recordings. Too many artifacts were found in the MEG recording of patient 15, because of previous surgery (meningioma), and FLAIR images were not available for patient 16. The EEG and MEG recordings of three other patients (patients 17–19) were rejected for source-imaging analysis because of poor interobserver agreement (kappa <0.40).

Figures 1 and 2 show examples of integrated EEG and MEG sources plotted on the MRI scans. Overall, MEG sources were significantly closer to the tubers (t test; p = 0.007). Figure 3 shows a scatter of 3D distances of the EEG and MEG sources to the closest tuber. The EEG source was closer to the tuber in only patient 1. The mean distance between source and tuber was 13.8 mm (range, 6.7– 29.3 mm) for MEG and 24.7 mm (range, 5.5–39.3 mm) for EEG. In two patients (patients 4 and 6) with only one source of interictal epileptiform activity on both EEG and MEG recordings, the nearest tuber to one modality differed from the other modality. The mean distance between tuber and source was still smaller for MEG (21.3 mm) than for EEG (26.4 mm). Interestingly, in patients with multiple localizations of interictal epileptiform activity, a good agreement was found between EEG and MEG sources regarding the closest tuber.

Figure 1.

Figure 1.

Example of source localization in patient 19. Integration of EEG (A) and magnetoencephalography (B) sources in 3D fluid-attenuated inversion recovery image. The crosshair indicates the tuber. The MUSIC result for epileptiform activity is shown in yellow. Note that the nearest tuber is the same tuber in both modalities.

Figure 1.

Figure 1.

Example of source localization in patient 19. Integration of EEG (A) and magnetoencephalography (B) sources in 3D fluid-attenuated inversion recovery image. The crosshair indicates the tuber. The MUSIC result for epileptiform activity is shown in yellow. Note that the nearest tuber is the same tuber in both modalities.

Figure 2.

Figure 2.

Example of source localization in patient 12. Integration of EEG (A) and magnetoencephalography (B) sources in 3D fluid-attenuated inversion recovery image. The crosshair indicates the tuber. The MUSIC result for epileptiform activity is shown in yellow. Note that the nearest tuber is a different tuber in both modalities.

Figure 2.

Figure 2.

Example of source localization in patient 12. Integration of EEG (A) and magnetoencephalography (B) sources in 3D fluid-attenuated inversion recovery image. The crosshair indicates the tuber. The MUSIC result for epileptiform activity is shown in yellow. Note that the nearest tuber is a different tuber in both modalities.

Figure 3.

Scatter of 3D distances of the EEG and magnetoencephalography sources to the closest tuber. Each number indicates a patient.

In agreement with a previous study (22), the epileptogenic tuber was most commonly (in nine patients) identified in the temporal region. In addition, tubers presumed to be epileptogenic were found parietally (three patients), occipitally (one patient), and frontally (one patient). Epileptogenic tubers were located on the right in 10 patients and on the left in four patients.


Three patients underwent epilepsy surgery (patients 1, 6, and 13). In patient 1, the ECoG showed epileptiform activity in the region posterior to the resected tuber. This was in agreement with MEG findings. In patient 6, ECoG recording showed epileptiform activity in the frontopolar and basal frontal regions. MEG spikes were found in the basal frontal region. In patient 13, ECoG recording showed epileptiform activity in the hippocampus and over the medial temporal gyrus, the latter in correspondence with MEG recording. Peri- or postsurgical complications were not seen. Patients 1 and 13 are seizure free (follow-up, 2 and 4 years). The seizure frequency in patient 6 improved by <50%. Epilepsy surgery was discussed in two more patients but postponed because of low seizure frequency.


We found that epileptiform activity detected with MEG is closer to a presumed epileptogenic tuber than is epileptiform activity detected with EEG. Although the value of MEG in the preoperative workup of candidates for epilepsy surgery has been acknowledged worldwide (23,24), experience with MEG in TSC patients is limited. Spike detection with MEG may be difficult and is subjective (19). For this reason, we analyzed only sources for which at least moderate interobserver agreement existed. Interobserver agreement tended to be worse when only a few spikes were detected.

MEG detected a single localization of epileptiform activity more often than EEG did. However, we cannot exclude the possibility that MEG was unable to detect radial sources, whereas EEG did. Although most of the analyzed traces were not recorded at the same time in all patients, only two patients had significantly more spikes in the MEG recording than in the EEG recording. Although interictal EEG recordings (sometimes HR-EEG) are used routinely in the preoperative workup of patients, our results show that spike detection is better with MEG recordings.

The characteristics of TSC lesions (e.g., multiple tubers) make it difficult to assess the association between the functional and the structural abnormalities. Tubers are often not far apart, and it is not surprising that EEG and MEG sources were not associated with the same tuber in two patients. We found that the distance from MEG source to EEG-associated tuber was smaller than the distance from EEG source to MEG-associated tuber. The difference between EEG and MEG recordings may not be as marked as one would expect from data in the literature. However, studies comparing EEG and MEG data often involved standard EEG recording, whereas HR-EEG, as used in our study, probably allows more accurate source localization. This would mean that a better agreement would be found between HR-EEG and MEG data.

It could be argued that the tuber closest to the epileptogenic source is not necessarily the epileptogenic tuber. Obviously immediate electrocorticography during epilepsy surgery or prolonged electrocorticography (ECoG) is the gold standard. MEG and video-EEG results have been proven to be equivalent in most patients considered for epilepsy surgery, with MEG providing additional information in a significant number of patients (25). The localization of interictal epileptiform activity has to be confirmed by ictal recordings before epilepsy surgery is performed. We are currently investigating the agreement between interictal and ictal source localization. If reproducible spikes are not detected during interictal EEG recording, or integration of the EEG source in the MRI does not identify a single tuber, simultaneous MEG recording is helpful to identify the tuber that should be resected. However, during and after resection of the presumed epileptogenic tuber, ECoG recordings should be performed for more accurate delineation of the irritative zone. Other techniques that can be used to identify the epileptogenic tuber include [11C]methyl-l-tryptophan ([11C] AMT) PET (26) and diffusion-weighted MRI (27). Both techniques can distinguish between epileptogenic and nonepileptogenic tubers.

Three of the patients underwent epilepsy surgery. In these patients, we found a found good agreement between interictal EEG, interictal MEG, and ictal recordings. Complete tuber resection took place in two patients, who remained seizure free postoperatively. In the third patient, only partial resection was possible because of the localization of the tuber in the eloquent cortex and the size of the tuber; unfortunately in this patient, seizure frequency was not changed in a clinically relevant way.

In conclusion, our results show that the epileptogenic source identified on MEG recordings was closer to the presumed epileptogenic tuber than was the source identified on EEG recordings from the same patients with TSC. In addition, interobserver agreement on source localization was greater with MEG recordings than with EEG recordings. Especially in patients with numerous tubers in whom interictal and ictal EEG recordings are taken for source localization, MEG source localization can help to delineate the source of the epileptiform activity and define the relation to the closest tubers. Hence, MEG is a useful technique and should be used to evaluate TSC patients considered for epilepsy surgery.


Acknowledgment:  F.E. Jansen was supported by the Epilepsy Fund of The Netherlands (grant 02-13) and by a research grant from the Jan Meerwaldt Stichting, The Netherlands. We thank Bob van Dijk, Jeroen Verbunt, Ilonka Manshanden, and Peterjan Ris at the KNAW MEG Center in Amsterdam for technical assistance, and Zeger Knops of the Department of Information and Computer Sciences, University of Utrecht, for matching the FLAIR and 3DT1 MRIs.