The patient gave her written informed consent and all procedures have been approved by the ethics committee of the University of Erlangen, Faculty of Medicine on 10.05.2011 (Ref. No. 4453).
Data was acquired from a 49-year-old female suffering from pharmaco-resistant focal onset epilepsy since the second year of life. The patient used 8 life-time antiepileptic drugs, but was still suffering from 100 to 200 seizures per month. The seizure semiology, involving tingling feeling at the right anterior torso, ascending feeling of nausea, then loss of consciousness and tonic or hypermotor movement of right arm and leg, was pointing to left frontocentral regions. Discordantly, diagnostic MRI revealed a right frontal FCD on a 3D-FLAIR sequence (with a resolution of 1 mm3) prior to source analysis. Morphometric MRI analysis (Huppertz et al. 2005) results were not very clear, but led to a suspicion of a possible second left frontal focal cortical dysplasia (Fig. 1b). However, visual reinspection of the MRI (3D-FLAIR at 3 T with voxels of 1 × 1 × 1 mm3) could not confirm this suspicion (Fig. 1a).
Interictal and ictal surface EEG (standard 10/20 montage) did not show reproducible epileptic discharges. FDG-PET failed to reveal a focal hypometabolism at the sites of the two suspected FCD, only left temporo-mesial structures showed a little less glucose uptake than contralateral. Because of non temporo-mesial seizure semiology this was regarded as unspecific. For intracranial EEG recordings each one depth electrode (AD-tech, Racine, WI U.S.A) was stereotactically inserted into the regions-of-interest derived from morphometric MRI-analysis (MRIcro (Chris Rorden, Version 1.37) imported into Iplan software, Brainlab, Feldkirchen, Germany; Wellmer et al. 2010) (Fig. 1c, d). The recording of strong interictal epileptic discharge activity confirmed the suspicion of FCD IIB in both localizations. Seizure onset, however, was documented only in the left hemispheric lesion—few seconds before clinical seizure onset (Fig. 1e). Responsibility of the left FCD IIB was further confirmed by the result of surgical treatment of the left, but not the right frontal FCD. Following stereotactic, lesion focused radiofrequency thermocoagulation the patient had a truncation of symptoms of her seizures (postoperatively only short arousal tonic and right arm but no more hypermotor seizure component). The most likely reason for failure to achieve complete seizure freedom is that the coagulation missed a small part of the lesion (for more details see discussion in Wellmer et al. 2016).
In addition to the diagnostic 3 T MRI, two sets of MRI data were acquired on two different 3 T scanners. The first scans were acquired prior to EEG/MEG source analysis for the main purpose of building the head model and finding cortical malformations. The second scan was done at a later date and was guided by the source analysis results (see Fig. 2 for a scheme of the analysis strategy). Note the radiological convention of left and right in all presented MR images (patient’s left is viewer’s right).
First Study MR Acquisition (Prior to Source Analysis)
A 3 T scanner (Gyroscan Intera/Achieva 3.0 T, System Release 2.5 (Philips Healthcare, Best, NL)) was used for the acquisition. The specific sequences were:
3D-T1-weighted (T1w) fast gradient-echo pulse sequence (TFE) using water selective excitation to avoid shifted fat signal (TR/TE/FA = 9.2/4.4 ms/9°, inversion prepulse every 1015.5 ms, cubic voxels of 1.17 mm edge length).
3D-T2w turbo spin echo pulse sequence (TR/TE = 2000/378 ms, cubic voxels, 1.17 mm edge length).
Diffusion tensor (DT) MRI using an echo planar imaging sequence (Stejskal-Tanner spin-echo, TR/TE = 7546/67 ms, cubic voxels, 1.875 mm edge length), with one volume with diffusion sensitivity b = 0 s/mm2 (i.e., flat diffusion gradient) and 20 volumes with b = 1000 s/mm2 in different directions, equally distributed on a sphere. Another volume with flat diffusion gradient, but with reversed spatial encoding gradients was acquired and used for susceptibility artifact correction (Ruthotto et al. 2012).
3D-FLAIR (Fluid-attenuated inversion recovery, TR/TE = 7000/322 ms, inversion time 2400 ms, cubic voxels, 1.17 mm edge length). This sequence was used to detect possible cortical malformations and lesions.
The acquisition times required for each of these four scans were approximately 7 min. In order to improve the co-registration between MRI and EEG/MEG electrodes/sensors, three gadolinium filled markers were placed on the nasion as well as inside the left and right ear canals prior to the MRI scan.
Second Study MR Acquisition (Guided by Source Analysis)
A second set of MRI data was acquired with another 3 T scanner (MAGNETOM Prisma 3.0 T, Release D13 [Siemens Medical Solutions, Erlangen, Germany]). The main reason for selecting another scanner for this part of the examination was to benefit from a novel MRI technique that employs localized excitation utilizing 2D selective RF pulses (Finsterbusch 2010) with parallel transmission (ZOOMit) (Blasche et al. 2012). Localized excitation allows to ‘zoom’ a field of view, restricting excitation to a desired area even within brain tissue without aliasing artifacts that occur when the FOV is smaller than the imaged object. This avoids the need to increase the number of phase encoding steps and the penalty of an increased minimum measurement time. The number of phase encoding steps necessary to obtain sufficient signal to noise can be used to increase the acquired volume in slice encoding direction, allowing a flexible definition of the volume of interest for searching the lesion in three dimensions. In the case presented here the acquired volume was a cuboid of 160 mm × 82 mm × 28 mm (lr × ap × fh, frequency × phase × slice encoding (2nd phase encoding)) with cubic voxels of 0.5 mm × 0.5 mm × 0.5 mm edge length.
Lesion visibility was additionally improved by the choice of contrast parameters. Our setting with (nominal) TR/TE/TI 2320/198/1800 ms in a 3D Turbo-Inversion Recovery technique with Flip angle control (SPACE) resulted in a combination of T2- and T1-weighting with sufficient signal strength to show the FCD. The acquisition time was about 13 min. Two different regions of interest (ROIs) at right frontal and left frontocentral locations were selected based on the findings of EMEG source analysis (explained in detail in the following sections).
EEG, MEG and ECG were recorded simultaneously in a magnetically shielded room. The EEG cap had 80 AgCl sintered ring electrodes (EASYCAP GmbH, Herrsching, Germany). The MEG was acquired with a whole head system with 275 axial gradiometers and 29 reference coils (OMEGA2005, CTF, VSM MedTech Ltd., Canada). The reference coils were used to calculate 3rd order synthetic gradiometers, thereby reducing the interference of magnetic fields originating from distant locations (e.g., Magnetocardiogram).
The patient was measured in supine position to reduce head movements and to avoid brain shift. Rice and colleagues (Rice et al. 2013) have shown that brain shift results in changes in CSF thickness and even these small changes affect EEG signals with 80% power difference on average due to the high conductivity of CSF.
The electrode positions were digitized with a Polhemus device (FASTRAK®, Polhemus Incorporated, Colchester, Vermont, U.S.A.) prior to the measurement. During the recordings the position of the head inside the MEG scanner was constantly measured via three head localization coils placed on the nasion and in the ear canals (same positions as the gadolinium markers in MRI).
In total seven runs were acquired. During the first run (7 min long) the median nerve of the right arm of the patient was stimulated with electrical pulses just above the motor threshold (used to calibrate skull conductivity for the head model, see (Aydin et al. 2014) for details). This run was followed by six 8 min long runs (2400 Hz sampling rate, low pass filtered at 600 Hz), in which the patient was advised to relax and close her eyes, aimed at measuring interictal epileptic discharges.
Calibrated Finite Element Head Model and Forward Solution
T1w and T2w MRIs were used to construct an individual seven-compartment head model that distinguishes scalp, skull spongiosa, skull compacta, dura mater, cerebrospinal fluid (CSF), gray matter (GM), and white matter (WM). The resulting segmentation and T1w MRI are shown in Fig. 3 upper and middle rows, respectively. The pipeline used for registration and segmentation was similar to the one explained before (Aydin et al. 2014), however, following recent findings (Ramon 2012; Ramon et al. 2014; Fiederer et al. 2015), an additional tissue type, the dura mater, was segmented and included as the seventh compartment. The segmentation of the dura mater, including the sagittal sinus space filled with venous blood that is surrounded by dura mater, was performed using Seg3DFootnote 1 and involved manual segmentation as well as some basic image processing steps such as smoothing and thresholding.
Diffusion tensors were calculated from the individual DTI data and used to model the WM conductivity tensors by an effective medium approach (Tuch et al. 2001; Rullmann et al. 2009; Ruthotto et al. 2012; Aydin et al. 2014). These conductivity tensors were later included into the head model to account for the anisotropic WM tissue.
The importance of skull conductivity has been shown for EEG and MEG source reconstruction in adults (Aydin et al. 2014) as well as for EEG in neonates (Roche-Labarbe et al. 2008). The conductivity of skull shows a high inter- and intra-individual variance. EEG source reconstructions are strongly influenced by changes in skull conductivity, while these effects are considerably smaller for MEG. Therefore, in (Aydin et al. 2014, see algorithm 2) we used a dipole scanning strategy that benefits from the different sensitivity profiles of EEG and MEG to calibrate skull conductivity. The calibration procedure used in this work could be summarized as first benefitting from the low sensitivity of MEG to skull conductivity and localizing the primary somatosensory cortex even for less suitable skull conductivity parameters. Then, fixing the location and determining the orientation of the dipole with EEG (to compensate for the insensitivity of MEG to quasi-radial source components). Finally, determining the appropriate skull conductivity by comparing the magnitudes of EEG and MEG dipoles for this fixed position and orientation. We have used the somatosensory P20/N20 response for the calibration procedure because it is well known that the generators of this component are localized in Brodmann area 3b and are focal, not too deep and mainly tangentially oriented (Allison et al. 1991). Details of the skull calibration procedure used in this study can be found in (Aydin et al. 2014). The calibrated conductivities were calculated as 0.0033 S/m for skull compacta and 0.0116 S/m for skull spongiosa. Other tissue conductivities (S/m) used in this study were: scalp (0.43) (Ramon et al. 2004), CSF (1.79) (Baumann et al. 1997), GM (0.33) (Fuchs et al. 1998), dura mater (0.1) (Ramon 2012).
A geometry adapted hexahedral finite element mesh was created out of the segmented MRI using SimBio-VGRIDFootnote 2. Geometry adapted hexahedral meshes provide a good balance by achieving better conformance to the geometry than regular hexahedral meshes whilst being less time-consuming and complicated than constructing tissue-surface based conforming tetrahedral meshes (Camacho et al. 1997; Wolters et al. 2007; Wagner et al. 2016).
The source space nodes were restricted to being located inside the GM without any orientation constraint. The source singularity was modeled with the Venant direct approach (Buchner et al. 1997; Wolters et al. 2007). To satisfy the Venant condition, for each source space node, it was checked whether the adjacent FE mesh nodes belong to elements which were labeled as GM (Vorwerk et al. 2012). The final source space had an average resolution of 2 mm (see blue points in Fig. 3 bottom row).
The finite element transfer matrix approach and the algebraic multigrid preconditioned conjugate gradient (AMG-CG) solver were used for increased computational efficiency (Wolters et al. 2002, 2004). The forward solution was calculated with piecewise trilinear basis functions using the SimBioFootnote 3 software.
Interictal Epileptic Discharges and Source Reconstruction
An experienced epileptologist (author SR) reviewed EEG and MEG traces separately, and marked 18 interictal epileptic discharges (IEDs). Eight of these IEDs were marked as EEG IEDs (activity more pronounced in EEG compared to MEG) with maximum negativity at F6 (will be called Espikes). Ten IEDs were marked as MEG only IEDs (Mspikes). Figure 4 shows averaged EEG and MEG signals as butterfly plots based on all IEDs (EMspikes) (top row), on Espikes only (middle row), and on Mspikes only (bottom row).
A current density approach, standardized low resolution brain electromagnetic tomography (sLORETA), was selected for source analysis (Pascual-Marqui 2002). sLORETA is a widely used source analysis method and it has been shown to perform well in situations in which multiple sources need to be accurately localized, which are temporally disentangled or whose leadfields are sufficiently uncorrelated (Pascual-Marqui 2002; Dümpelmann et al. 2012; Lucka et al. 2012). The leadfield matrices calculated with SimBio and source space points calculated with custom written Matlab code were imported into the CURRY 7Footnote 4 software in order to solve the inverse problem.
After performing eddy current and susceptibility corrections by following the procedure explained elsewhere (Ruthotto et al. 2012; Aydin et al. 2014), the FSL-BEDPOSTX routine was used to calculate the distribution of diffusion parameters at each voxel using Markov Chain Monte Carlo sampling. Afterwards, the FSL-PROBTRACKX function was used to perform probabilistic tractography between two ROIs (Behrens et al. 2007). These ROIs were selected based on EMEG source analysis and the subsequent ZOOMit data.