1 Multimodal Functional Imaging Lab, Biomedical Engineering Department, McGill University, Montréal, Québec, Canada.
2 Integrated Program in Neuroscience, McGill University, Montréal, Québec, Canada.
3 Montreal Neurological Institute, Department of Neurology and Neurosurgery, McGill University, Montréal, Québec, Canada.
4 Aix Marseille University, INSERM, INS, Inst Neurosci Syst, Marseille, France.
5 APHM, Timone Hospital, Epileptology and Cerebral Rhythmology, Marseille, France.
6 Physnum Team, Centre De Recherches Mathématiques, Montréal, Québec, Canada.
7 Electrical Engineering Department, École De Technologie Supérieure, Montréal, Québec, Canada.
8 Multimodal Functional Imaging Lab, Department of Physics and Concordia School of Health, Concordia University, Montréal, Québec, Canada.

Due to the ill-posed nature of source imaging, EEG/MEG source localization remains challenging, particularly for localizing deep brain activity and resting-state signals with low signal-to-noise ratios. Functional connectivity estimated by EEG/MEG is further affected by source-leakage, a spatial blurring effect that complicates interpretation. Validation is therefore essential before applying EEG/MEG source imaging in clinical contexts. Using simultaneous MEG and stereotactic EEG (SEEG) recordings from patients with focal epilepsy, we present a comprehensive validation of MEG source imaging of epileptic discharges, resting-state oscillations, and connectivity. Virtual SEEG was computed from MEG source maps to enable direct quantitative comparison with in situ SEEG. Using the Maximum Entropy on the Mean method, MEG localized the generators of epileptic spikes with a median error of 15 ± 12 mm, with deep generators showing larger errors. MEG-derived resting-state power showed significant spatial correlations with SEEG, with frequency-specific correspondence, stronger in the alpha and beta bands than in the theta band. MEG functional connectomes estimated using leakage-corrected amplitude envelope correlation aligned relatively well with SEEG, whereas connectomes from the leakage-corrected phase-based metric were inaccurate. These findings delineate conditions under which MEG source imaging reliably captures epileptic and resting-state activity, even with low signal-to-noise ratios, and conditions when it is unreliable.