Abstracts

Rationale: Due to the ill-posed nature of EEG/MEG source imaging, the accuracy of EEG/MEG estimated sources requires validation, before considering localization of epileptic discharges and resting-state analysis for clinical applications. We recently validated MEG source imaging with an intracerebral EEG atlas of physiological brain activity (Frauscher Brain 2018) at a group level in terms of resting-state oscillations (Afnan NIMG 2023) and connectivity (Afnan in prep). In this study, we aimed to validate MEG source imaging of oscillations and connectivity patterns at a single-subject level using simultaneous MEG and stereotaxic EEG (SEEG).


Methods: We considered simultaneous MEG and SEEG acquired from 5 patients with epilepsy, acquired at the MEG center of the Institut de Neurosciences des Systèmes (Aix Marseille University, France) (Pizzo Nature Comm 2019). For each patient, a 1-min segment with oscillations in a dominant frequency band was marked in MEG, where no artifact or epileptic discharge was visibly present. MEG inverse problem was solved using wavelet Maximum Entropy on the Mean method (Lina IEEE TMBE 2012; Afnan HBM 2024). To quantitatively compare MEM results with SEEG, we projected MEG sources at each SEEG channel position using an SEEG forward model (Grova HBM 2016) (Fig. 1 for an example patient). We computed the relative power spectral density (PSD) for both MEG and SEEG for each channel. The spatial correlation between two modalities was calculated for the average PSD in the frequency band of interest (Fig.1). For connectivity analysis, we computed pairwise connectivity for all channel pairs using Amplitude Envelope Correlation (AEC) and orthogonalized AEC (OAEC). We compared the spatial correlation between MEG and SEEG connectomes for each metric.


Results: Fig.2 shows the cross-modal correlation between MEG and SEEG for the example patient in Fig.1 for PSD (Fig. 2A-D) and connectivity (Fig. 2E). PSD: For patient 1 (Fig1, 2), a correlation of ~0.54 was found for average PSD. When the correlation was computed for superficial and deep channels based on an eccentricity threshold of 60 mm (a measure of source depth, with low values indicating deep channels), the cross-modal correlations for superficial and deep channels were 0.68 and 0.18, respectively. For the 5 patients, the cross-modal correlations for average PSD were 0.31 ± 0.12 (0.42 ± 0.12 for superficial and 0.15 ± 0.16 for deep channels). Connectivity: We found moderate correlations between MEG and SEEG connectomes for AEC. In contrast, for OAEC, a metric that corrects zero-lag connectivity, the correlation between MEG and SEEG connectomes decreased.


Conclusions: MEG retrieved accurately oscillatory and connectivity patterns when compared to SEEG, whereas more accuracy was found for superficial activity, reflecting our group-level findings with non-simultaneous data (Afnan NIMG 2023). For connectivity, the decrease in cross-modal correlations found by metric corrected for zero-lag connectivity highlights a trade-off: although MEG may capture more connectivity due to source leakage, removing zero-lag connectivity also eliminates true connections, leading to a decrease in cross-modal correlation.


Funding: NSERC, CIHR, FRQNT.