Myelin-informed forward models for M/EEG source reconstruction

Saskia Helbling,Sofie S Meyer,Nikolaus Weiskopf
DOI: https://doi.org/10.1101/2024.06.30.601378
2024-07-02
Abstract:Magnetoencephalography (MEG) and Electroencephalography (EEG) provide direct electrophysiological measures at an excellent temporal resolution, but the spatial resolution of source-reconstructed current activity is limited to several millimetres. Here we show, using simulations of MEG signals and Bayesian model comparison, that non-invasive myelin estimates from high-resolution quantitative magnetic resonance imaging (MRI) can enhance MEG/EEG source reconstruction. Our approach assumes that MEG/EEG signals primarily arise from the synchronised activity of pyramidal cells, and since most of the myelin in the cortical sheet originates from these cells, myelin density can predict the strength of cortical sources measured by MEG/EEG. Leveraging recent advances in quantitative MRI, we exploit this structure-function relationship and scale the leadfields of the forward model according to the local myelin density estimates from in vivo quantitative MRI to inform MEG/EEG source reconstruction. Using Bayesian model comparison and dipole localisation errors (DLEs), we demonstrate that adapting local forward fields to reflect increased local myelination at the site of a simulated source explains the simulated data better than models without such leadfield scaling. Our model comparison framework proves sensitive to myelin changes in simulations with exact coregistration and moderate-to-high sensor-level signal-to-noise ratios (≥10 dB) for the multiple sparse priors (MSP) and empirical Bayesian beamformer (EBB) approaches. Furthermore, we sought to infer the microstructure giving rise to specific functional activation patterns by comparing the myelin-informed model which was used to generate the activation with a set of test forward models incorporating different myelination patterns. We found that the direction of myelin changes, however not their magnitude, can be inferred by Bayesian model comparison. Finally, we apply myelin-informed forward models to MEG data from a visuo-motor experiment. We demonstrate improved source reconstruction accuracy using myelin estimates from a quantitative longitudinal relaxation (R1) map and discuss the limitations of our approach.
Neuroscience
What problem does this paper attempt to address?