F123. Ultra-high-density scalp EEG outperforms localized invasive ECoG grids in inferring depth of seizure foci

Abstract

Introduction During pre-surgical evaluation for medically intractable epilepsy, intracranial electrocorticography (ECoG) is often used to locate seizure foci. However, ECoG grids usually have small spatial coverage because large coverage requires large craniotomies. This may pose difficulties in inferring the depth of the focus, which is important because foci can lie deep inside the brain, e.g. in insular, mesial or basal regions, or within a sulcus. Our team recently demonstrated theoretically [Grover & Venkatesh, Proc. IEEE, ‘17] and experimentally[Robinson et al., Scientific Reports ‘17] that ultra-high-density (UHD) scalp EEG (with sub-centimeter electrode spacing) can recover high spatial resolution information of brain activity. Here, we ask whether UHD-EEG with full scalp coverage can outperform spatially localized ECoG grids in inferring the depth of seizure foci. Methods In our simulation study, ECoG and UHD-EEG electrodes were placed on a template-head MRI. Brainstorm and OpenMEEG were used to generate EEG and ECoG forward models using the Boundary Element Method for the ICBM-152 Brain template with 15,765 dipoles. For 20-electrode ECoG simulations, 8 different locations were considered (over frontal, fronto-parietal, posterior-temporal and occipital regions on left and right hemispheres), with sources at different locations in each case. For 40-electrode ECoG simulations, two adjacent 20-electrode ECoG grids were considered, again for varying source locations. A single radially oriented dipole was manually activated at different depths from each of the 8 locations on the cortex, beneath the 20-electrode ECoG grid. For EEG, 128, 256 and 1378 electrodes (the last being a theoretical UHD EEG system) with whole scalp coverage were considered. Noiseless ECoG and EEG recordings were simulated, and source localization was performed to estimate the source location using the L2-regularized MNE algorithm. Results of source localization were then compared using two standard metrics: the distance of the center of the reconstruction from the true source (bias) and the diffusive extent of the reconstruction (width of Point Spread Function, PSF). Results The reconstruction-bias using UHD-EEG is consistently about 2 cm less than any ECoG reconstruction. PSF widths of deeper sources are observed to be slightly larger for ultra-high-density EEG than other modalities. Nevertheless, visual inference of depth is more accurate using reconstructions of ultra-high-density EEG. Conclusion UHD-EEG can complement other modalities used during pre-surgical evaluation for epilepsy, before and after intracranial electrode implantation, particularly for inferring the presence of a deep seizure focus. ECoG recordings strongly bias reconstructions towards the cortical surface. Future work will similarly compare stereo-EEG and UHD-EEG, for inferring depth of seizure foci, and evaluate how UHD-EEG can inform stereo-EEG electrode placement.