Authors :
First Author: Jawata Afnan, M.Sc – McGill University
Presenting Author: Christophe Grova, PhD – Concordia University, McGill University
Zhengchen Cai, PhD – Postdoctoral fellow, Department of Neurology and Neurosurgery, Montreal Neurological Institute, McGill University; Jean-Marc Lina, PhD – Professor, Electrical Engineering Department, École De Technologie Supérieure, Montréal; Chifaou Abdallah, M.D – PhD student, Integrated Program in Neuroscience, McGill University; Tamir Avigdor, M.Sc – PhD candidate, Integrated Program in Neuroscience, McGill University; Édouard Delaire, M.Sc. – PhD candidate, Department of Physics, Concordia University; Tanguy Hedrich, PhD – Biomedical Engineering Department – Multimodal Functional Imaging Lab, McGill University; Nicolás von Ellenrieder, PhD – Department of Neurology and Neurosurgery – Montreal Neurological Institute, McGill University; Eliane Kobayashi, PhD – Associate Professor, Montreal Neurological Institute, McGill University; Birgit Frauscher, MD, PhD – Associate Professor, Department of Neurology, Montreal Neurological Institute, McGill University; Jean Gotman, PhD – Professor, Montreal Neurological Institute, McGill University; Christophe Grova, PhD – Associate Professor, Department of Physics, Concordia University
Rationale:
EEG/MEG source imaging (EMSI) of epileptic activity from deep generators, including the hippocampus, is often challenging due to the higher sensitivity of EEG/MEG to superficial regions and to the spatial configuration of subcortical structures. We demonstrated the ability of the Maximum Entropy on the Mean (MEM) to accurately localize the superficial cortical generators and their spatial extent, when applied either in the time domain (cMEM) (Chowdhury et al Neuroimage 2016) or in the wavelet domain, wavelet- MEM (wMEM) for localizing oscillations (Afnan et al Neuroimage 2023). Here, we propose depth-weighted adaptations of MEM (dcMEM, dwMEM) to localize deep generators more accurately. These methods were evaluated using realistic MEG/EEG simulations of epileptic activity and actual EEG/MEG recordings from patients with focal epilepsy.
Methods:
We incorporated depth-weighting within the MEM framework, following our proposed strategy to reconstruct functional Near-InfraRed Spectroscopy data (Cai et al Sc. Report 2022), compensating for the bias toward superficial generators. We also included a mesh of both hippocampi, as an additional deep structure in the source model. Evaluation (Fig.1A) We generated 5400 realistic simulations of interictal epileptic discharges for MEG and High-Density EEG (HD-EEG) involving three levels of spatial extent of the generators and three levels of signal-to-noise ratio (SNR), before investigating EMSI on clinical HD-EEG data in 16 patients and clinical MEG data in 9 patients. Interictal epileptic discharges were marked by visual inspection in MEG and HD-EEG (C. A and B.F). We applied four EMSI methods: cMEM, wMEM,,the newly proposed dcMEM, and dwMEM. The ground truth was defined as the true simulated generator and as a drawn region of interest based on clinical information available for patients (anatomical MRI, neurophysiological evaluation, or intracranial EEG findings). Performance of EMSI results was evaluated using three metrics: Area Under the ROC Curve (AUC) (measuring spatial overlap), Spatial Dispersion (SD) (measuring spatial spread), and minimum distance localization error (D
min).
Results:
Fig.1B shows the differences in accuracy metrics between cMEM and dcMEM, and wMEM and dwMEM, for 300 MEG simulations (spatial extent=3, SNR=2) covering the whole brain. For deep sources, depth-weighted MEM improved the localization, as reflected by increased AUC and decreased SD and Dmin. Depth-weighted MEMs did not deteriorate localization accuracy for superficial regions (see AUC and Dmin). For patients’ data, we observed improvement in localization for deep sources, especially for the patients with mesial temporal epilepsy, for which standard cMEM and wMEM failed to reconstruct the generator in the hippocampus while the depth-weighted implementation recovered hippocampal activity during the rising phase of the spike (see example in Figure 2).
Conclusions:
Depth-weighted extension of cMEM and wMEM improves the localization of deep sources without deteriorating the localization of the superficial sources. This was demonstrated using extensive simulations with MEG and EEG and clinical MEG and HD-EEG recordings for epilepsy patients.
Funding: NSERC, CIHR, and FRQNT.