Abstracts

Rationale: The hemispheric dominance of language networks is crucial in the presurgical evaluation of patients with drug-resistant focal epilepsy. Several methods exist to map task-related evoked responses to the cerebral cortex and estimate hemispheric laterality indices (LIs). However, there is no consensus on optimal methodologies, and it is unclear whether control tasks improve estimates of language dominance. This study aimed to compare language dominance classifications derived from different MEG response mapping and LI estimation pipelines to those established using fMRI.


Methods: MEG data from 72 temporal lobe epilepsy patients (mean age: 39.19 ± 11.96) in the Epilepsy Connectome Project were analyzed. Recordings were acquired using MEGIN 306-channel MEG system. Patients performed a visual semantic decision task (animal names) and a control task (pseudo-letter strings). MEG data were preprocessed with spatiotemporal filtering and MEGnet automated ICA classification, then epoched into task and control trials from -0.3 to 2s relative to stimulus onset. Sensor-level evoked responses were mapped to the cortex using LCMV, weighted minimum norms, and dSPM. Response time series were estimated for four regions of interest (angular, frontal, temporal, lateral). LIs were calculated for both task (Animal) and task-control (Animal-Symbol) conditions over 300 ms windows advanced in 10 ms steps. LIs were derived using source-magnitude, vertex counts with a half-maximal threshold, and a weighted-threshold bootstrap approach. Time-resolved regional LIs from MEG were validated against those from an fMRI approach based on an analogous auditory task. Concordance of ternary language dominance classifications (left, right, symmetric) between MEG and fMRI was estimated.


Results: Source estimation using LCMV and LI calculation via the bootstrap method consistently yielded the highest concordance between MEG and fMRI. The task-control (Animal-Symbol) differential response significantly improved MEG-fMRI concordance. In the temporal ROI, the Animal-Symbol contrast yielded a maximal MEG-fMRI concordance of 81.94% for the interval 0.23-0.53s using LCMV and the bootstrap approach (Fig. 1c). For the Animal condition alone, the maximal concordance was 68.44% (0.24-0.54s), increasing to 69.44% using optimal response intervals for individual patients. In the lateral ROI, maximal MEG-fMRI concordance was 73.61% (0.23-0.53s) with LCMV and the bootstrap method, improving to 80.55% using optimal intervals for individual patients. For the Animal condition alone, concordance reached 65.27% (0.24-0.54s) and did not improve with individual optimization.


Conclusions: Our findings suggest that LCMV source estimation and a weighted-threshold bootstrap approach for LI calculation may offer significant advantages for determining language dominance from MEG evoked responses. Differential responses relative to a control task enhance the mapping of language networks using MEG. Predictors of MEG-fMRI discordances in hemispheric dominance classifications warrant further investigation

Funding: Supported by the Epilepsy Connectome Project (U01NS093650; NIH) & MEGIN US LLC (FP00024550).