Cortex-anchored sensor-space harmonics for event-related EEG

Abstract

Scalp event-related potentials (ERPs) measured with electroencephalography (EEG) are temporally precise but spatially bandwidth-limited: skull and scalp blur cortical activity, and event-related EEG is still usually represented in electrode coordinates or dataset-specific data-driven components rather than in a sensor-space coordinate system linked to cortical anatomy. Here we introduce a cortex-anchored sensor-space basis obtained by forward-projecting cortical Laplace-Beltrami (LB) eigenmodes through a realistic EEG head model, yielding a multiscale dictionary whose ordering follows cortical spatial frequency. Using ERP-CORE (7 paradigms, 39 participants), we benchmarked the forward-projected LB basis against (i) spherical harmonics defined on the same montage and (ii) group PCA/ICA bases learned from trial-averaged time-frequency (TF) maps. Across canonical components (N170, N2pc, N400, P3b, LRP, ERN), we quantified reconstruction efficiency (R2 as a function of the number of modes), concentration of evoked TF energy across modes, split-half reliability of mode scores (ICC), and low-dimensional reconstruction of group-level ERP contrast topographies. LB closely matched spherical harmonics in R2, but concentrated evoked TF energy more strongly in low-order modes: for N170, N400, P3b, and ERN, the first 10 LB modes captured about 70% of normalized TF energy, whereas spherical harmonics typically required 15-18 modes. Low-to-mid LB mode scores showed moderate-to-excellent reliability, often comparable to or slightly exceeding spherical harmonics. In addition, 10- 15 LB modes reconstructed canonical ERP contrast maps with high correlations while preserving expected sensor-space organization (posterior N170, centro-parietal N400/P3, fronto-central ERN). These results show that forward-mapped LB eigenmodes provide a compact, anatomy-linked sensor representation for event-related EEG that complements spherical and data-adaptive bases and offers a reusable feature space for geometryinformed neural signal analysis.