Gregor Strobbe

Influence of skull modeling approaches on EEG source localization

Electroencephalographic source localization (ESL) relies on an accurate model representing the human head for the computation of the forward solution. In this head model, the skull is of utmost importance due to its complex geometry and low conductivity compared to the other tissues inside the head. We investigated the influence of using different skull modeling approaches on ESL. These approaches, consisting in skull conductivity and geometry modeling simplifications, make use of X-ray computed tomography (CT) and magnetic resonance (MR) images to generate seven different head models. A head model with an accurately segmented skull from CT images, including spongy and compact bone compartments as well as some air-filled cavities, was used as the reference model. EEG simulations were performed for a configuration of 32 and 128 electrodes, and for both noiseless and noisy data. The results show that skull geometry simplifications have a larger effect on ESL than those of the conductivity modeling. This suggests that accurate skull modeling is important in order to achieve reliable results for ESL that are useful in a clinical environment. We recommend the following guidelines to be taken into account for skull modeling in the generation of subject-specific head models: (i) If CT images are available, i.e., if the geometry of the skull and its different tissue types can be accurately segmented, the conductivity should be modeled as isotropic heterogeneous. The spongy bone might be segmented as an erosion of the compact bone; (ii) when only MR images are available, the skull base should be represented as accurately as possible and the conductivity can be modeled as isotropic heterogeneous, segmenting the spongy bone directly from the MR image; (iii) a large number of EEG electrodes should be used to obtain high spatial sampling, which reduces the localization errors at realistic noise levels.

Bayesian model selection of template forward models for EEG source reconstruction.

Several EEG source reconstruction techniques have been proposed to identify the generating neuronal sources of electrical activity measured on the scalp. The solution of these techniques depends directly on the accuracy of the forward model that is inverted. Recently, a parametric empirical Bayesian (PEB) framework for distributed source reconstruction in EEG/MEG was introduced and implemented in the Statistical Parametric Mapping (SPM) software. The framework allows us to compare different forward modeling approaches, using real data, instead of using more traditional simulated data from an assumed true forward model. In the absence of a subject specific MR image, a 3-layered boundary element method (BEM) template head model is currently used including a scalp, skull and brain compartment. In this study, we introduced volumetric template head models based on the finite difference method (FDM). We constructed a FDM head model equivalent to the BEM model and an extended FDM model including CSF. These models were compared within the context of three different types of source priors related to the type of inversion used in the PEB framework: independent and identically distributed (IID) sources, equivalent to classical minimum norm approaches, coherence (COH) priors similar to methods such as LORETA, and multiple sparse priors (MSP). The resulting models were compared based on ERP data of 20 subjects using Bayesian model selection for group studies. The reconstructed activity was also compared with the findings of previous studies using functional magnetic resonance imaging. We found very strong evidence in favor of the extended FDM head model with CSF and assuming MSP. These results suggest that the use of realistic volumetric forward models can improve PEB EEG source reconstruction.

Seizure Onset Zone Localization from Ictal High-Density EEG in Refractory Focal Epilepsy

Epilepsy surgery is the most efficient treatment option for patients with refractory epilepsy. Before surgery, it is of utmost importance to accurately delineate the seizure onset zone (SOZ). Non-invasive EEG is the most used neuroimaging technique to diagnose epilepsy, but it is hard to localize the SOZ from EEG due to its low spatial resolution and because epilepsy is a network disease, with several brain regions becoming active during a seizure. In this work, we propose and validate an approach based on EEG source imaging (ESI) combined with functional connectivity analysis to overcome these problems. We considered both simulations and real data of patients. Ictal epochs of 204-channel EEG and subsets down to 32 channels were analyzed. ESI was done using realistic head models and LORETA was used as inverse technique. The connectivity pattern between the reconstructed sources was calculated, and the source with the highest number of outgoing connections was selected as SOZ. We compared this algorithm with a more straightforward approach, i.e. selecting the source with the highest power after ESI as the SOZ. We found that functional connectivity analysis estimated the SOZ consistently closer to the simulated EZ/RZ than localization based on maximal power. Performance, however, decreased when 128 electrodes or less were used, especially in the realistic data. The results show the added value of functional connectivity analysis for SOZ localization, when the EEG is obtained with a high-density setup. Next to this, the method can potentially be used as objective tool in clinical settings.

Electrical source imaging of interictal spikes using multiple sparse volumetric priors for presurgical epileptogenic focus localization

Electrical source imaging of interictal spikes observed in EEG recordings of patients with refractory epilepsy provides useful information to localize the epileptogenic focus during the presurgical evaluation. However, the selection of the time points or time epochs of the spikes in order to estimate the origin of the activity remains a challenge. In this study, we consider a Bayesian EEG source imaging technique for distributed sources, i.e. the multiple volumetric sparse priors (MSVP) approach. The approach allows to estimate the time courses of the intensity of the sources corresponding with a specific time epoch of the spike. Based on presurgical averaged interictal spikes in six patients who were successfully treated with surgery, we estimated the time courses of the source intensities for three different time epochs: (i) an epoch starting 50 ms before the spike peak and ending at 50% of the spike peak during the rising phase of the spike, (ii) an epoch starting 50 ms before the spike peak and ending at the spike peak and (iii) an epoch containing the full spike time period starting 50 ms before the spike peak and ending 230 ms after the spike peak. To identify the primary source of the spike activity, the source with the maximum energy from 50 ms before the spike peak till 50% of the spike peak was subsequently selected for each of the time windows. For comparison, the activity at the spike peaks and at 50% of the peaks was localized using the LORETA inversion technique and an ECD approach. Both patient-specific spherical forward models and patient-specific 5-layered finite difference models were considered to evaluate the influence of the forward model. Based on the resected zones in each of the patients, extracted from post-operative MR images, we compared the distances to the resection border of the estimated activity. Using the spherical models, the distances to the resection border for the MSVP approach and each of the different time epochs were in the same range as the LORETA and ECD techniques. We found distances smaller than 23 mm, with robust results for all the patients. For the finite difference models, we found that the distances to the resection border for the MSVP inversions of the full spike time epochs were generally smaller compared to the MSVP inversions of the time epochs before the spike peak. The results also suggest that the inversions using the finite difference models resulted in slightly smaller distances to the resection border compared to the spherical models. The results we obtained are promising because the MSVP approach allows to study the network of the estimated source-intensities and allows to characterize the spatial extent of the underlying sources.

Multiple sparse volumetric priors for distributed EEG source reconstruction.

We revisit the multiple sparse priors (MSP) algorithm implemented in the statistical parametric mapping software (SPM) for distributed EEG source reconstruction (Friston et al., 2008). In the present implementation, multiple cortical patches are introduced as source priors based on a dipole source space restricted to a cortical surface mesh. In this note, we present a technique to construct volumetric cortical regions to introduce as source priors by restricting the dipole source space to a segmented gray matter layer and using a region growing approach. This extension allows to reconstruct brain structures besides the cortical surface and facilitates the use of more realistic volumetric head models including more layers, such as cerebrospinal fluid (CSF), compared to the standard 3-layered scalp-skull-brain head models. We illustrated the technique with ERP data and anatomical MR images in 12 subjects. Based on the segmented gray matter for each of the subjects, cortical regions were created and introduced as source priors for MSP-inversion assuming two types of head models. The standard 3-layered scalp-skull-brain head models and extended 4-layered head models including CSF. We compared these models with the current implementation by assessing the free energy corresponding with each of the reconstructions using Bayesian model selection for group studies. Strong evidence was found in favor of the volumetric MSP approach compared to the MSP approach based on cortical patches for both types of head models. Overall, the strongest evidence was found in favor of the volumetric MSP reconstructions based on the extended head models including CSF. These results were verified by comparing the reconstructed activity. The use of volumetric cortical regions as source priors is a useful complement to the present implementation as it allows to introduce more complex head models and volumetric source priors in future studies.

The Role of Skull Modeling in EEG Source Imaging for Patients with Refractory Temporal Lobe Epilepsy

We investigated the influence of different skull modeling approaches on EEG source imaging (ESI), using data of six patients with refractory temporal lobe epilepsy who later underwent successful epilepsy surgery. Four realistic head models with different skull compartments, based on finite difference methods, were constructed for each patient: (i) Three models had skulls with compact and spongy bone compartments as well as air-filled cavities, segmented from either computed tomography (CT), magnetic resonance imaging (MRI) or a CT-template and (ii) one model included a MRI-based skull with a single compact bone compartment. In all patients we performed ESI of single and averaged spikes marked in the clinical 27-channel EEG by the epileptologist. To analyze at which time point the dipole estimations were closer to the resected zone, ESI was performed at two time instants: the half-rising phase and peak of the spike. The estimated sources for each model were validated against the resected area, as indicated by the postoperative MRI. Our results showed that single spike analysis was highly influenced by the signal-to-noise ratio (SNR), yielding estimations with smaller distances to the resected volume at the peak of the spike. Although averaging reduced the SNR effects, it did not always result in dipole estimations lying closer to the resection. The proposed skull modeling approaches did not lead to significant differences in the localization of the irritative zone from clinical EEG data with low spatial sampling density. Furthermore, we showed that a simple skull model (MRI-based) resulted in similar accuracy in dipole estimation compared to more complex head models (based on CT- or CT-template). Therefore, all the considered head models can be used in the presurgical evaluation of patients with temporal lobe epilepsy to localize the irritative zone from low-density clinical EEG recordings.

EEG source connectivity to localize the seizure onset zone in patients with drug resistant epilepsy

Electrical source imaging (ESI) from interictal scalp EEG is increasingly validated and used as a valuable tool in the presurgical evaluation of epilepsy as a reflection of the irritative zone. ESI of ictal scalp EEG to localize the seizure onset zone (SOZ) remains challenging. We investigated the value of an approach for ictal imaging using ESI and functional connectivity analysis (FC). Ictal scalp EEG from 111 seizures in 27 patients who had Engel class I outcome at least 1 year following resective surgery was analyzed. For every seizure, an artifact-free epoch close to the seizure onset was selected and ESI using LORETA was applied. In addition, the reconstructed sources underwent FC using the spectrum-weighted Adaptive Directed Transfer Function. This resulted in the estimation of the SOZ in two ways: (i) the source with maximal power after ESI, (ii) the source with the strongest outgoing connections after combined ESI and FC. Next, we calculated the distance between the estimated SOZ and the border of the resected zone (RZ) for both approaches and called this the localization error ((i) LEpow and (ii) LEconn respectively). By comparing LEpow and LEconn, we assessed the added value of FC. The source with maximal power after ESI was inside the RZ (LEpow = 0 mm) in 31% of the seizures and estimated within 10 mm from the border of the RZ (LEpow ≤ 10 mm) in 42%. Using ESI and FC, these numbers increased to 72% for LEconn = 0 mm and 94% for LEconn ≤ 10 mm. FC provided a significant added value to ESI alone (p < 0.001). ESI combined with subsequent FC is able to localize the SOZ in a non-invasive way with high accuracy. Therefore it could be a valuable tool in the presurgical evaluation of epilepsy.

Improved Localization of Seizure Onset Zones Using Spatiotemporal Constraints and Time-Varying Source Connectivity

Presurgical evaluation of brain neural activity is commonly carried out in refractory epilepsy patients to delineate as accurately as possible the seizure onset zone (SOZ) before epilepsy surgery. In practice, any subjective interpretation of electroencephalographic (EEG) recordings is hindered mainly because of the highly stochastic behavior of the epileptic activity. We propose a new method for dynamic source connectivity analysis that aims to accurately localize the seizure onset zones by explicitly including temporal, spectral, and spatial information of the brain neural activity extracted from EEG recordings. In particular, we encode the source nonstationarities in three critical stages of processing: Inverse problem solution, estimation of the time courses extracted from the regions of interest, and connectivity assessment. With the aim to correctly encode all temporal dynamics of the seizure-related neural network, a directed functional connectivity measure is employed to quantify the information flow variations over the time window of interest. Obtained results on simulated and real EEG data confirm that the proposed approach improves the accuracy of SOZ localization.

EEG beamforming to extract better features of motor imagery in a two-class real-time BCI

A general problem in the design of an EEG-BCI system is the poor quality and low robustness of the extracted features, affecting overall performance. However, BCI systems that are applicable in real-time and outside clinical settings require high performance. Therefore, we have to improve the current methods for feature extraction. In this work, we investigated EEG source reconstruction techniques to enhance the extracted features based on a linearly constrained minimum variance (LCMV) beamformer. Beamformers allow for easy incorporation of anatomical data and are applicable in real-time. A 32-channel EEG-BCI system was designed for a two-class motor imagery (MI) paradigm. We optimized a synchronous system for two untrained subjects and investigated two aspects. First, we investigated the effect of using beamformers calculated on the basis of three different head models: a template 3-layered boundary element method (BEM) head model, a 3-layered personalized BEM head model and a personalized 5-layered finite difference method (FDM) head model including white and gray matter, CSF, scalp and skull tissue. Second, we investigated the influence of how the regions of interest, areas of expected MI activity, were constructed. On the one hand, they were chosen around electrodes C3 and C4, as hand MI activity theoretically is expected here. On the other hand, they were constructed based on the actual activated regions identified by an fMRI scan. Subsequently, an asynchronous system was derived for one of the subjects and an optimal balance between speed and accuracy was found. Lastly, a real-time application was made. These systems were evaluated by their accuracy, defined as the percentage of correct left and right classifications. From the real-time application, the information transfer rate (ITR) was also determined. An accuracy of 86.60 ± 4.40% was achieved for subject 1 and 78.71 ± 0.73% for subject 2. This gives an average accuracy of 82.66 ± 2.57%. We found that the use of a personalized FDM model improved the accuracy of the system, on average 24.22% with respect to the template BEM model and on average 5.15% with respect to the personalized BEM model. Including fMRI spatial priors did not improve accuracy. Personal fine- tuning largely resolved the robustness problems arising due to the differences in head geometry and neurophysiology between subjects. A real-time average accuracy of 64.26% was reached and the maximum ITR was 6.71 bits/min. We conclude that beamformers calculated with a personalized FDM model have great potential to ameliorate feature extraction and, as a consequence, to improve the performance of real-time BCI systems.Observability of electrical potentials from deep brain sources to surface EEG remains unclear and debated among the neuroscience community. This question is particularly crucial in the temporal lobe epilepsies investigations because they involve complex (mesial and/or lateral) epileptogenic networks (Maillard et al., 2004; Bartolomei et al, 2008). At present, when mesial structures are supposed to be epileptogenic only clinical indirect evidences are used to diagnose mesial temporal lobe (MTL) epilepsy. Based on this methodology and on drug resistance evidence, surgical treatment can be proposed without the need of invasive intracerebral investigation. Reported results of this surgery demonstrate an incomplete success (70-80%; McIntosh et al. 2012) which indicate that indirect evidences of the contribution of mesial sources are not sufficient. Seven patients undergoing pre-surgical evaluation of drug resistant epilepsy were selected from a prospective series of twenty eight patients in whom simultaneous depth and surface EEG recordings had been performed since 2009. Above these patients, three had right temporal lobe (TLE) epilepsy and four left TLE. Simultaneous SEEG-EEG signals were recorded using 128 channels placed on the same acquisition system that avoids the need to synchronize both signals. Intracerebral interictal spikes (IIS) were selected on depth EEG signals blinded to EEG signals. These IIS were triggered as temporally known (T0) brain sources due to their specific waveform and the high signal to noise ratio. Then, after IIS characterization and classification, EEG signals were automatically averaged according to the T0 markers. Averaged EEG signals were finally characterized (3D mapping, duration, amplitude and statistics) and clustered using hierarchical clustering method. Overview of the data collection and analysis process is presented in figure 1. In mean in our population, 9 depth EEG electrodes and 16 surface EEG electrodes were simultaneously used. 684±186 IIS were selected by patient for a total number of spikes in our population of 4787. According to the anatomical distribution of the IIS, 21 foci were defined and classified according to three categories: mesial (limbic structures plus collateral fissure; M, 9 foci), mesial and neocortical (M+NC, 5 foci) and neocortical part of the temporal lobe (NC, 7 foci). Comparison between SEEG spikes and averaged EEG spikes on the most activated electrode at T0 was presented in table 1. Concerning 3D Map amplitude, negative pole were always seen in the temporo-basal region for both M, M+NC and NC foci and positive pole were only observed for M+NC and NC foci. Using Walsh statistical test, 8 EEG channels in mean was presented averaged amplitude at t0 statistically different of the averaged background activity. Three different clusters were fund using the hierarchical clustering method on averaged EEG signals: 1) all patients included in the M foci class and 2) all patients included in the M+NC and NC foci class and 3) one patient with an atypical brain source. Observability of deep sources with surface EEG recordings is possible. Electrical sources from mesial temporal lobe cannot be considered as closed electrical field structures. The main problem to observe signals from these deep structures concern the signal to noise ratio. Indeed, spontaneous surface spikes originated from mesial structures cannot be seen without averaging. Hierarchical clustering method and 3D map amplitude of average EEG signals at t0 seems to indicate that M contributions was different to M+NC and NC contributions. So ICA method associated with a predetermined topography constraint should detect (without the need of simultaneous depth EEG) the mesial contribution in raw EEG signals.

Comparison of fMRI activation and EEG source localization using beamformers during motor response in the Stroop task: Preliminary results

The simultaneous measurement of neuro-functional MRI and EEG provides the opportunity to investigate both haemodynamic and electrical activity in the human brain non-invasively. This multimodal technique makes it possible to combine the advantages of EEG (millisecond temporal resolution) with those of fMRI (millimeter spatial accuracy). However, the combination of EEG and fMRI suffers also from their limitations: there is no clear relationship between neuronal activity, the EEG and the fMRI signals; mismatches are observed between EEG and fMRI analyses and there are experimental limitations. In this study we present a qualitative analysis of simultaneous acquired EEG/fMRI data from a single subject performing the Stroop task. The EEG data and the fMRI time series are processed separately focused on the motor responses. We apply a beamformer spatial filter to the EEG data to localize the electrical activity corresponding to the motor responses of the right and the left hand. The areas of maximum electrical activation are compared with the fMRI activation clusters. Both analyses show activation around the primary motor cortex. These results are an indication to start with a more sophisticated and integrated analysis of simultaneous acquired EEG/fMRI data during the Stroop task.