EEG 'Default-Mode' Marker Fails as Unitary Biomarker in 96-Subject Study

A 15 October 2026 NeuroImage study of 96 scalp EEG recordings finds that the widely used 'EEG-DMN-like' sensor-level pattern does not replicate as a unitary marker. Salvador Herrera-Pérez and colleagues at the (affiliations per the paper) show that posterior alpha suppression and frontal theta enhancement hold across three internally oriented tasks, while connectivity effects collapse depending on which estimator family and analysis window the researcher selects.

The study appeared on 15 October 2026 as article 122204 in Volume 340 of NeuroImage, published open access under a Creative Commons licence (DOI: 10.1016/j.neuroimage.2026.122204).

What did the researchers measure?

The primary cohort was OpenNeuro dataset ds004148: 60 subjects recorded with 61-channel scalp EEG during three internally oriented tasks — MATH, MEMORY and MUSIC. An independent external validation cohort, EEGMAT from PhysioNet, contributed 36 subjects with 19-channel EEG during mental arithmetic.

The team computed spectral, topographic and connectivity metrics under a per-subject theta-localised 60-second primary window, plus a fixed central 60-second sensitivity window. Six connectivity estimators covered phase-lag-tolerant, zero-lag-suppressing and amplitude-envelope families. Replication required the expected direction of effect, Cohen's dz of at least 0.2, and a 95 percent bootstrap confidence interval excluding zero. The authors also ran reference-scheme sensitivity analyses using the original online reference and REST.

Which components replicated?

Posterior alpha suppression and frontal theta enhancement were directionally consistent across all three ds004148 tasks, with effect sizes of dz = 0.21–0.47 and topographically coherent posterior alpha clusters. The spectral signature, in plain terms, survived every task and validation test for direction.

Connectivity told a different story. Lag-tolerant estimators reached small-to-moderate effects in MEMORY and MUSIC (dz approximately 0.20–0.43) but attenuated in MATH. Zero-lag-suppressing estimators failed the replication criterion across every cohort and task cell.

How sensitive were the results?

The choice of analysis window mattered. Under the central 60-second window, MATH alpha and alpha-1 effects fell to near zero, and MEMORY frontal theta dropped below the dz threshold. The external EEGMAT cohort confirmed the direction of the primary effects but not their magnitude.

Reference scheme and montage density changed little. Direction was preserved under montage harmonisation and under both alternative reference schemes, with zero or 16 sign changes under REST and only near-threshold per-metric attenuations.

What does this change for EEG practice?

The authors conclude that the sensor-level EEG-DMN-like pattern is best interpreted as a proxy of the rest-to-internally-oriented-cognition state transition, not as a unitary biomarker of default-mode network activity. Reproducibility is component-specific, window-specific and estimator-specific: the spectral signature is directionally robust across cohorts, references and montage densities, whereas connectivity depends critically on estimator family.

The practical instruction is explicit: future studies using this sensor-level proxy must pre-specify the estimator family and the task window. For reading rooms and quantitative EEG labs, that means a reported alpha-band fronto-posterior coupling value cannot be compared across studies unless both parameters match. Raw data remain openly available from OpenNeuro (ds004148) and PhysioNet (EEGMAT), so groups can test their own pipelines against these thresholds directly.