Abstracts

Rationale: Recent research emphasizes the cerebellum’s pivotal role in seizure networks, including disrupted functional and structural connectivity, volume alterations, perfusion changes, and instances of lesional cerebellar epilepsy. However, assessing cerebellar electrophysiology remains challenging due to unreliable forward modeling caused by its intricate folding. To address these challenges, we developed ARCUS¹, a method for automatic reconstruction and segmentation of the cerebellum using clinically applicable MRI (1mm thickness) from individual subjects. ARCUS enables feasible forward modeling and precise source estimation from the cerebellum.


Methods: We consecutively applied ARCUS to patients with focal epilepsy for validation and investigated the general cerebellar signal-to-noise ratio (SNR) maps obtained from SQUID-MEG and EEG. Patients who underwent MEG scans for presurgical evaluation of epilepsy at the Martinos Center between May 2021, and May 2022 were included. MEG signals were recorded using a 306-channel MEG system (Elekta Neuromag, VecterView) with gradiometers and magnetometers while patients were comfortably supine. A 70-channel EEG cap, appropriate for the head size, was used. The forward model incorporated the cerebellum into the source space using a three-layer BEM method. Cortical surfaces were reconstructed from T1 structural data, resulting in about 100,000 vertices per cerebral hemisphere and a similar number for the cerebellum. The forward model was used to calculate the MEG/EEG signal based on neural activity modeled as current dipoles aligned with the vertex surface normals. We adopted the SNR definition of Goldenholz et al. (2009) for MEG/EEG SNR analysis. Differential SNR maps were computed by subtracting SNR values of each modality (magnetometer, gradiometer, and EEG).


Results: MEG was obtained from 118 patients (2.1–66 years, average 26.8 years; 57 female). EEG was not collected in 9 patients. After excluding 12 patients due to large lesions and 3 missing MRIs, Freesurfer reconstruction was incomplete in 19 patients due to the artifacts (e.g., VNS, dental work, motion degradation). The three-layer BEM model could not be created in 29 patients due to layer crossings. Overall, 53 patients were included for further investigation. Among them, 51 showed reliable cerebellar segmentation based on visual inspection by a trained neurosurgeon (T.M). Median SNR maps on the template cerebellum (Figure 1) revealed higher SNR in the anterior lobe for magnetometers compared to gradiometers, while the posterior lobe showed higher SNR for EEG than magnetometers. Box plots (Figure 2) illustrated SNR difference across segmentation.


Conclusions: This study verifies the ARCUS algorithm for cerebellar reconstruction and segmentation in a large patient population, revealing distinct SNR patterns between modalities. Our comprehensive investigation provides valuable insights into cerebellar sources in MEG/EEG SNR mappings, potentially guiding researchers and clinicians in selecting the optimal modality for cerebellar studies.


Funding: Overseas Research Fellow of Japan Society for the Promotion of Science
(Reference) 1. Samuelsson JG, et al. bioRxiv. 2020:2020.11. 30.405522.