Context-dependent PSIICOS: A novel framework for functional connectivity estimation accounting for task-related power leakage

Abstract

Functional connectivity (FC) analysis using non-invasive neuroimaging methods, such as MEG and EEG, is often confounded by artifacts from spatial leakage and task-related power modulations. To address these limitations, we present Context-Dependent PSIICOS (CD-PSIICOS), a novel framework that improves the estimation of FC by incorporating task-specific cortical power distributions into the projection operator applied to the vectorized sensor-space cross-spectrum. Unlike the original PSIICOS (Phase Shift Invariant Imaging of Coherent Sources) approach, designed to suppress spatial leakage from all the sources, CD-PSIICOS dynamically adjusts the projection based on the active source distribution, enabling more accurate suppression of spatial leakage while preserving true zero-phase interactions. We validated CD-PSIICOS using realistic simulations and a multi-subject MEG dataset. The results demonstrate that CD-PSIICOS outperforms the original PSIICOS in suppressing artifacts at the lower projection ranks, maintaining robust detection of functional networks across theta and gamma frequency bands. By requiring lower projection ranks for optimal performance, CD-PSIICOS facilitates the reconstruction of physiologically relevant networks with improved sensitivity and stability.