Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Rising Stars | Poster Sessions | Data Blitz

Reproducibility Crisis in Functional Connectivity Analyses of EEG and MEG: An Approach to Reduce False Connectivity Findings

Poster Session A - Saturday, March 7, 2026, 3:00 – 5:00 pm PST, Fairview/Kitsilano Ballrooms

Anthony Herdman¹ (aherdman@audiospeech.ubc.ca); ¹University of British Columbia

The reproducibility crisis in neuroscience has exposed fundamental limitations in functional connectivity (FC) analyses of electroencephalography and magnetoencephalography data. FC aims to quantify interactions among brain regions but is confounded by signal mixing and spatial spread, producing spurious zero-lag phase relationships and inflated false positive FC findings. Source reconstruction using inverse solutions partially mitigates these effects, but source leakage remains pervasive and can generate false connectivity even between distant regions. We evaluated how different inverse modeling approaches affect FC accuracy when ground truth is known. Using the SimMEEG framework, realistic EEG datasets were simulated with known source connectivity. Inverse modeling was performed using common single-source inverse solutions (e.g., LCMV, MNE, SLORETA) and a multisource beamformer designed to reduce source mixing. Localization and FC performances of inverse solutions were assessed using multiple metrics, including source localization and orientation errors, waveform reconstruction error, and FC detection accuracy. Because true source-level FC connectivity were simulated and thus known, true-positive and false-positive FC detection rates were compared across inverse solutions. Results showed that multisource beamforming consistently outperformed all single-source inverse solutions. It achieved very high localization accuracy with minimal false-positive localizations. It also had the highest sensitivity and specificity for FC detection. In contrast, single-source inverse solutions exhibited substantial source leakage and high false-positive FC detections, even when non-zero-lag measures were applied. These findings demonstrate that unresolved source leakage is a critical contributor to finding false FCs and that multisource beamforming offers a potential solution for improving reliability and reproducibility among FC studies.

Topic Area: METHODS: Neuroimaging