Authors :
Presenting Author: Teppei Matsubara, MD, PhD – Massachusetts General Hospital
Abbas Sohrabpour, PhD – Massachusetts General Hospital
Seppo Ahlfors, PhD – Massachusetts General Hospital
Padmavathi Sundaram, PhD – Massachusetts General Hospital
Steven Stufflebeam, MD – A. A. Martinos Center for Biomedical Imaging
Rationale:
Recent research highlights the cerebellum’s pivotal role in seizure networks, including disrupted connectivity, volumetric changes, altered perfusion, and cases of lesional cerebellar epilepsy. Yet, cerebellar electrophysiology remains difficult to assess—largely due to the challenges of forward modeling caused by its complex folding. To address this, we developed ARCUS1, an automated method for cerebellar reconstruction and segmentation using clinically acquired 1 mm MRI. ARCUS enables region-specific forward modeling from individual patients, making cerebellar source analysis feasible.
Concurrently, optically pumped magnetometers (OPMs) are emerging as a powerful MEG technology. By being placed directly on the scalp—like EEG—OPMs minimize source-to-sensor distance and are expected to enhance SNR, especially for deeper or posterior brain structures such as the cerebellum.
In this study, we applied ARCUS to compare cerebellar signal detectability across EEG, SQUID-based MEG, and multiple OPM configurations in a real-world clinical cohort.
Methods:
We analyzed patients with intractable epilepsy who underwent presurgical SQUID-MEG and MRI during 2021–2022. ARCUS was used for regional cerebellar segmentation. Forward modeling was performed using a 3-layer BEM, with dipoles oriented perpendicular to the cortical surface to reflect pyramidal neurons in the cerebrum and Purkinje cells in the cerebellum. Dipole amplitude was fixed at 10 nAm, and noise estimates were derived from clinical EEG and MEG recordings and uniformly applied across modalities. SNR was calculated at each vertex and averaged across 16 cerebellar regions.
Sensor configurations included standard EEG and SQUID-MEG, as well as three OPM layouts: OPM(eeg) (aligned to EEG electrodes), OPM(squid) (projected from SQUID positions), and OPM(grid) (a 4×5 posterior grid anchored to the inion). SNR differences were computed between modalities and correlated with anatomical features (e.g., angle to skull surface) and patient age.
Results:
Among 63 patients (mean age 27.1 ± 14.3 years, range 4.7–61.4), ARCUS segmentation failed in only 3 cases—demonstrating high robustness across varying MRI quality and clinical conditions. SNR maps showed SQUID-MEG provided strong sensitivity in superficial posterior cerebellar regions, while EEG was more uniform but weaker (Fig.1A, B). The OPM(grid) layout outperformed SQUID-MEG in posterior regions, including Crus I/II and the inferior vermis (Fig.1C, D).
Region-specific correlations revealed age-dependent shifts in SNR difference for SQUID-MEG vs EEG and OPM(grid), but not OPM(eeg) (Fig.2 left). A positive correlation between source angle and SNR difference was observed in MEG and OPM configurations, consistent with orientation sensitivity (Fig.2 right).
Conclusions:
This modeling-based analysis reveals distinct modality-dependent patterns in cerebellar signal detectability and supports the potential of OPM systems to enhance posterior fossa coverage. When paired with automated segmentation via ARCUS, this framework provides a scalable, individualized approach for cerebellar MEG modeling in patients with epilepsy.
1. Samuelsson JG, et al. bioRxiv. 2020:2020.11. 30.405522.
Funding: None