Introduction
Despite considerable advancements in electroencephalography (EEG) and magnetoencephalography (MEG) source analysis, these techniques’ sensitivity, specificity, and spatial resolution to completely replace invasive recordings are still under discussion. In many cases, source analysis is used only to guide placement of depth EEG-electrodes. Although knowing the exact region to implement invasive electrodes is very critical, the ultimate aim of EEG/MEG source analysis is to minimize the necessity for invasive recordings. This could lead to important benefits by avoiding the complications of invasive recording procedures (Hamer et al.
2002; Wellmer et al.
2012). Furthermore, invasive recordings can only measure activity within a close distance from the sensors, suffer from low spatial sampling due to limited numbers of invasive electrodes, and exhibit a tunnel view effect due to limited coverage (Lüders et al.
2006).
Similar to invasive recordings, EEG and MEG have high temporal resolution in the range of milliseconds (ms) and measure the electrical activity of neurons directly without using indirect phenomena like hemodynamics or metabolism, which is the case for functional magnetic resonance imaging (fMRI) or positron emission tomography (PET). However, the spatial resolutions of EEG and MEG are lower in comparison to fMRI and their sensitivities decrease with the distance between sources and sensors. EEG and MEG source analysis bears some specific challenges. Some of these are related to the forward problem: developing methodology and pipelines to model the head and the brain as accurate as possible, while still keeping the setup time and computational costs at a reasonable level for clinical use. Although there has been considerable advancement in this area, especially within the finite element framework (Wolters et al.
2002; Rullmann et al.
2009; Vorwerk et al.
2014), three compartment models calculated with the boundary element method are still the most widely deployed ones in the field. The main difficulty related to the inverse problem of EEG and MEG is its non-uniqueness. This means that there is, without further prior information, an infinite number of source configurations that results in the same EEG/MEG signals (Hämäläinen et al.
1993). Many promising inverse approaches have been developed to alleviate this problem by using different constraints (Pascual-Marqui
2002; Lucka et al.
2012; Chowdhury et al.
2013; Lina et al.
2014). Instead of solving the EEG and MEG inverse problems independently, performing combined EEG/MEG (EMEG) source analysis leads to significant improvements (Aydin et al.
2015; Chowdhury et al.
2015). These improvements are particularly significant for scenarios with low signal-to-noise ratio (SNR), such as for deep sources or at the spike onset (Aydin et al.
2015).
The specificity of EEG, MEG and EMEG source analysis could be significantly increased by incorporating other available information. The additional information might come from other functional imaging techniques such as fMRI or PET, from seizure semiology, or from MRI sequences sensitive to structural changes and lesions. In general, not every lesion evident in structural MRI is related to the epilepsy, but some types of lesions, such as focal cortical dysplasia (FCD) type IIB are shown to be highly epileptogenic (Wagner et al.
2011). Furthermore, it has been reported that in up to 73% of MRI negative cases, histology shows an underlying FCD (Lee et al.
2005). Therefore, reinvestigating structural MRI by incorporating the findings from source analysis may be a beneficial practice in presurgical epilepsy diagnosis (Moore et al.
2002; Wang et al.
2012).
In this study, we introduce a pipeline that combines information from EMEG source analysis, seizure semiology, and high resolution structural MRI in presurgical epilepsy diagnosis. We constructed a high resolution (1 mm edge length) head model that distinguishes seven different tissue types, uses diffusion tensor imaging (DTI) to amend the anisotropic white matter compartment, and benefits from a calibration procedure to estimate individual skull conductivity. This head model was then used to solve the forward problem with the finite element method (Aydin et al.
2014) and perform EMEG source analysis (Aydin et al.
2015). The most important novelty of this paper is coupling EMEG source analysis with a ‘zoomed’ MRI sequence that allows localized excitation utilizing parallel transmission (ZOOMit) (Blasche et al.
2012). This technique is capable of acquiring data with 0.5 mm voxel edge length of a restricted area within a reasonable time. By combining EMEG source analysis, seizure semiology information, and the ZOOMit MRI, a subtle FCD, which was undetectable at the lower resolution (1 mm), was detected near the epileptic focus localized by EMEG. To the best of our knowledge this is the first study that combines source analysis and zoomed MRI in the field of epilepsy.
Discussion
We studied a new multimodal approach in presurgical epilepsy diagnosis that benefits from (i) combined information from EEG and MEG, (ii) an individual high resolution finite element head model, (iii) individually calibrated skull conductivity, and (iv) recent advancements in morphometric MRI analysis and multi transmit and receive head coils. The first step was simultaneous EEG/MEG recordings followed by an MRI session acquiring T1w, T2w, DTI and FLAIR data with typical resolutions (1.875 mm edge length for DTI and 1.17 mm for the rest). This first study MRI session and the combined somatosensory evoked potential and field data were used to construct the calibrated finite element head model and thus solving the forward problem of EEG/MEG. Combined EEG/MEG (EMEG) source analysis was then performed using a distributed source approach to calculate the active areas in the brain, close to the peak of the averaged interictal epileptic discharges, as well as at earlier phases. Later, a second MRI session was performed, this time, using a new zooming technique (ZOOMit) to acquire high resolution images (0.5 mm voxel edge length) within two limited regions. These regions were selected based on EMEG source analysis near the peak (right frontal region) and at an earlier phase (left frontocentral region). The ZOOMit MRI revealed one relatively clear FCD at the right frontal region, and another subtle FCD at the left frontocentral region. Of interest, the left frontocentral FCD was not identifiable in 3D-FLAIR and only this one, and not the right frontal FCD, was concordant with seizure semiology. The second FCD was not detected in visual evaluation of any previous clinical MRIs at 3 T acquired and investigated in different centers; not even retrospectively. DTI tractography suggested a possible anatomical pathway supporting a fast propagation from the left frontocentral to the right frontal FCD. Further converging evidence for the hypothesis, although not very clear, was obtained from the morphometric MRI analysis (Huppertz et al.
2005) following an epilepsy specific protocol (Wellmer et al.
2013). The morphometric analysis hinted at a suspicious area, among others, close to left frontocentral FCD. Based on this converging evidence only the left frontocentral FCD had been treated using stereotactic radiofrequency thermocoagulation and the surgical outcome as well as the intracranial EEG supported our diagnosis (see Fig.
1 and the “
Patient” section).
In addition to the cluster in the left hemisphere, due to quite clear further activity in MEG and EEG (see Mspikes topographies in Figs.
4,
5), we also found a cluster in the right hemisphere at −23 ms. This cluster was not investigated further because it was not in concordance with the seizure semiology. However, it is possible that the right central region was involved in interictal spikes as well but not in seizure generation. Although, irritative and seizure onset zones usually coincide it has been shown that this might not always be the case. For example, for some patients with bi-temporal spikes (with irritative zones in both right and left hemispheres) seizure freedom was achieved after performing operation in just one of the temporal lobes (Lüders et al.
2006). This also points to the importance of not using just one type of information but performing a multimodal strategy in presurgical epilepsy diagnosis.
As mentioned in the “
Introduction” section there have been significant advances in the field of EEG/MEG source analysis in recent years, and multimodal combination of other noninvasive techniques, such as MRI, with source analysis could provide a promising way to increase the specificity. FCDs are intrinsically epileptogenic cortical malformations, resection of which leads to a high chance of seizure freedom (Sisodiya
2000). It has been reported that even in many MRI negative cases, post-operative histology could show an underlying FCD: (Lee et al.
2005) showed this number could be as high as 73%. In (McGonigal et al.
2007) histopathology showed FCD or hippocampal sclerosis for 12 out of 23 MRI negative patients, this number was 9 out of 29 in (Bien et al.
2009). Therefore, there is no doubt that many patients will benefit if the number of false negative MRIs could be reduced and the most obvious way to do that is going for higher spatial resolutions and better SNR, while keeping the examination time short enough to avoid patient movement. The main advantage of the here proposed ZOOMit technique was that it benefitted from localized excitation utilizing 2D selective RF pulses (Finsterbusch
2010) with parallel transmission (Blasche et al.
2012). As mentioned in the methods section 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. This localized excitation combined with fine tuned contrast parameters allowed us to obtain a combination of T2- and T1-weighting with increased lesion visibility and high resolution using a 3 T MRI.
FCDs can be difficult to identify with MRI and vice versa, ambiguous structural alterations may falsely be regarded as FCDs. Especially subtle findings may therefore not always imply significance for seizure generation. As we have shown regarding the right frontal FCD in this patient, the epileptic focus responsible for the seizures might even be far away from it. It is possible that the patient studied here has two potential epileptic foci and the right frontal one is pharmacosensitive; therefore, we did not record seizures from there. The study of (Brodbeck et al.
2010) showed the importance of source localization in MRI negative cases and (Zhang et al.
2014) presented an important review on the increasing value of multimodal imaging in epilepsy.
Although indirectly, in this study we have also demonstrated the importance of EMEG source analysis for the planning of intracranial electrode placement. In this study, if the intracranial electrodes had been placed near the right frontal FCD alone, the earlier epileptic activity arising from the left frontocentral FCD could not have been measured (because invasive electrodes are only sensitive to activity from close proximity). This would probably have led to the implantation of a second set of intracranial electrodes, based on seizure semiology, in order to detect the left frontocentral focus. This indicates the importance of tailoring the implantation of invasive electrodes by combining the information obtained from noninvasive EEG/MEG, MRI and the seizure semiology. In this direction (Knowlton et al.
2009) and (Agirre-Arrizubieta et al.
2014) have also shown the importance of MEG in the placement of intracranial EEG.
Another important point was that only combined EEG/MEG source analysis was able to localize the activity at the epileptic focus (activity near the left frontocentral FCD). Even though single modality EEG or MEG source analysis also detected activity very similar to the EMEG near the peak of the spike (localization of the right frontal FCD), their results were far away from the epileptic focus, which was only detectable near the spike onset. The topographies in Figs.
4 and
5 show hints of a left central dipolar pattern for both EEG and MEG. However, the SNR at this time instant was considerably too low and we think this is the reason for seeing the left frontocentral localization only in EMEG. This result is in line with previous studies showing the advantages of combined EEG/MEG in comparison to single modality source analysis because of the more stable source reconstructions and the superior spatial resolution (Cohen and Cuffin
1987; Fuchs et al.
1998; Baillet et al.
1999; Huang et al.
2007; Aydin et al.
2014,
2015; Chowdhury et al.
2015; Lucka
2015). In one of our previous studies we could show that the complementary information of EEG and MEG is especially important when the SNR of the data is low, such as at the spike onset (Aydin et al.
2015). In (Aydin et al.
2015) and (Chowdhury et al.
2015) it was also shown that the EMEG localizations were not simply the union of EEG and MEG results but a rather complicated interplay of both modalities compensating their relative shortcomings.
In this study, based on the results of our previous work (Aydin et al.
2015), we used an advanced realistic head model with seven different tissue compartments and white matter anisotropy modeled from DTI. However, we are aware that the generation and use of such a realistic head model might often not be feasible in clinical routine work and its clinical value will have to be evaluated in a larger series of patients. In such cases, following the findings of (Aydin et al.
2014; Vorwerk et al.
2014), we would suggest calibrating the skull conductivity and adding CSF and white/gray matter distinction which requires overall less effort, but still could considerably improve the results, and could especially enable combined EEG and MEG source analysis.
The main aim of this work was the presentation of a proof-of-principle, i.e., the methodology for a new multimodal presurgical epilepsy diagnosis approach and its feasibility and success in a case study with a multi-focal epilepsy patient that suffered from pharmaco-resistant focal onset epilepsy for 47 years of her life. As an outlook, the most important future goal will now be the reproduction of the presented results in a study with a larger group of epilepsy patients, a goal, which might not only be tackled by our working group, since methodology for combined EEG/MEG source analysis and MRI scanners using parallel transmit technology are now becoming more and more available.