Predictive regression modeling with MEG/EEG: from source power to signals and cognitive states

Abstract

Predicting biomedical outcomes from Magnetoencephalography and Electroencephalography (M/EEG) is central to applications like decoding, brain-computer-interfaces (BCI) or biomarker development and is facilitated by supervised machine learning. Yet most of the literature is concerned with within-subject classification. Here, we focus on predicting continuous outcomes from M/EEG signal power across subjects. Considering different generative mechanisms for M/EEG signals and the biomedical outcome, we propose statistically-consistent predictive models that avoid sourcereconstruction based on the covariance as representation. Our mathematical analysis and ground truth simulations demonstrated that consistent parameter estimation can be obtained with Source Power Comodulation (SPoC) supervised spatial filtering or by embedding with Riemannian geometry. Additional simulations revealed that Riemannian methods were more robust to model violations, in particular geometric distortions induced by individual anatomy. To estimate the relative contribution of brain dynamics and anatomy to prediction performance, we propose a novel model inspection procedure based on biophysical forward modeling. Applied to cross-subject prediction, the analysis revealed that the Riemannian model better exploited anatomical information while sensitivity to brain dynamics was similar across methods. We then probed the robustness of the models across different data cleaning options. Environmental denoising was globally important but Riemannian models were strikingly robust and continued performing even without preprocessing. Our results suggest each method has its niche: SPoC is practical for within-subject prediction while the Riemannian model may enable simple end-to-end learning.

Movement Decoding using Spatio-Spectral Features of Cortical and Subcortical Local Field Potentials

“…The rest of the filters are obtained by solving a generalized eigenvalue decomposition problem [14]. As shown in [13], the matrix W spoc recovers the inverse of mixing matrix A defined in (1).…”

“…The most common approach for movement decoding from intracranial recordings is based on the extraction of band-power features at specific frequency ranges, such as in the beta ( [13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Hz) and gamma ([30-100] Hz) bands, after band-pass filtering the signal. This strategy has been applied by several studies using either subcortical STN-LFPs [7,8] or cortical ECoG signals [11,14].…”

“…Brain recordings are in most of the cases the consequence of electrical activity of neural populations sources firing in the proximity of the electrode. Unfortunately, due to the overlapping activity of multiple sources, such neural populations cannot be observed directly and must be inferred with uncertainty from their distorted recording representations [24]. Statistical generative models assume that brain signals arise from activity of uncorrelated sources, and that such sources appear geometrically distorted in the signal recording as a consequence of a linear mixing transformation, as illustrated in Figure 1.…”

“…Importantly, due to the overlapping activity of multiple sources, specific neural population activity cannot be observed directly and must be inferred with uncertainty [13]. Statistical generative models often assume that brain signals arise from activity of uncorrelated sources, and that such sources appear distorted in the signal recording as a consequence of a linear mixing (due to volume conduction), as illustrated in Figure 1.…”

Decoding subjective emotional arousal from EEG during an immersive Virtual Reality experience

“…SPoC is a supervised regression approach, in which a target variable (here: subjective emotional arousal) guides the extraction of relevant M/EEG oscillatory components (here: alpha power). SPoC has been used to predict single-trial reaction times from alpha power in a hand motor task (Meinel et al, 2016), muscular contraction from beta power (Sabbagh et al, 2020), and difficulty levels of a video game from theta and alpha power (Naumann et al, 2016). CSP is used to decompose a multivariate signal into components that maximize the difference in variance between distinct classes (here: periods of high and low emotional arousal).…”