Abstracts

Rationale: Noninvasive neuroimaging methods like electroencephalography (EEG) and magnetoencephalography (MEG) are key in localizing the epileptogenic zone in children with drug-resistant epilepsy (DRE). Yet, their ability to detect deep brain sources remains unclear due to anatomical and technical limitations. Here, we developed a realistic, MRI-based 3D-printed pediatric head phantom, incorporating deep dipolar sources to evaluate electric and magnetic source imaging (ESI and MSI) under varying signal-to-noise ratio (SNR) levels. Our aim is to examine whether ESI/MSI can reliably localize deep brain sources. We hypothesize that both EEG and MEG are capable of localizing deep sources with a localization error (LE) of ~20 mm, assuming that proper data analysis methods are used.

Methods: A realistic three-layer head phantom, mimicking the human head geometry and electrical properties, was constructed (Fig. 1A). The phantom incorporated implanted dipoles in regions of interest (i.e. thalamus, brainstem, insula, amygdala, and orbital gyrus; Fig. 1B). We stimulated the dipoles with an epoch containing evident interictal spikes obtained from stereotactic EEG recordings of a DRE patient’s hippocampus (Fig. 1C) and performed MEG and EEG recordings of the phantom. Then we superimposed different levels of background brain activity obtained from MEG and EEG recordings of a typically developing child to the phantom recordings to obtain different SNR levels (-5, 0, 5, 10, 15 dB, and no noise) (Fig. 1D). ESI and MSI were evaluated using dipole scanning (single, clustered, and averaged dipoles) and dynamic statistical parametric mapping (dSPM; single and averaged maps) across varying SNRs. LE was calculated as the Euclidean distance between the estimated and implanted source location.

Results: All deep sources were localized by both MEG and EEG, with accuracy improving as SNR increased. In ESI dipole scanning, brainstem showed the lowest LE (9 mm with averaged dipoles; Fig. 2A), and amygdala errors were below 25 mm. MSI single dipoles localized the orbital gyrus at 14 ± 7 mm error, and brainstem errors ranged from 11-13 mm across all SNRs. For the insula, MSI averaged dipoles showed low LE (16 mm, no noise conditions; Fig. 2A). For dSPM, ESI averaged maps achieved optimal localization for amygdala (12 mm; Fig. 2A) and orbital gyri (~14-17 mm), but ESI single dSPM maps frequently failed to localize deeper sources. MSI averaged dSPM maps provided better localization (e.g., Orbital gyrus 2 at 13 mm; Fig. 2A), though MSI single dSPM maps also struggled with deep sources. No statistically significant comparisons were found between dipole scanning and dSPM methods across modalities after correcting for multiple comparisons (Fig. 2B-C).

Conclusions: Both MEG and EEG localized deep sources with errors of 19±5 mm and 22±6 mm, respectively. Averaging improved dSPM-based localization, especially at low SNRs. Both SNR and source depth were key factors affecting accuracy. Our findings show evidence that the synergy of the two modalities may improve the presurgical planning for pediatric DRE by guiding the selection of optimal methods to achieve higher localization accuracy.

Funding: R01NS104116-01A1 by NINDS.