Non-unitary reproducibility of an EEG-DMN-like sensor-level proxy: metric-family, task-window and paradigm effects
Salvador Herrera-Pérez1, Vanessa Moscardó2, Patricia López Mases3
1Facultad de Ciencias de la Salud. Universidad Internacional de Valencia, Spain; Centro de Estudios en Ciencia de Datos e Inteligencia Artificial ESenCIA, Universidad Internacional de Valencia, Spain.
Background:
Posterior α suppression, frontal θ enhancement and fronto-posterior α-band coupling are routinely interpreted together as an 'EEG-DMN-like' rest-to-task marker. Whether they replicate as a unitary pattern or dissociate across metric families, task windows and cognitive paradigms remains insufficiently characterised within a single harmonised pipeline.
Methods:
ds004148 (n = 60; 61-channel EEG; three internally oriented tasks: MATH, MEMORY and MUSIC) was the primary cohort, and EEGMAT (n = 36; 19-channel EEG; mental arithmetic) the independent external validation cohort. Spectral, topographic and connectivity metrics were computed under a per-subject θ-localised 60-s primary window with a fixed central 60-s sensitivity window. Six connectivity estimators covering phase-lag-tolerant, zero-lag-suppressing and amplitude-envelope families were computed. Replication required the expected direction, Cohen's dz ≥ 0.2 and a 95 % bootstrap CI excluding zero. Reference-scheme sensitivity analyses (original online reference and REST) were also performed.
Results:
Posterior α suppression and frontal θ enhancement were directionally consistent across the three ds004148 tasks (dz = 0.21-0.47) with topographically coherent posterior α clusters. Under the central window, MATH α/α₁ fell to near zero and MEMORY frontal θ dropped below threshold. Lag-tolerant connectivity reached small-to-moderate effects in MEMORY and MUSIC (dz ≈ 0.20-0.43) and attenuated in MATH; zero-lag-suppressing estimators failed the replication criterion across cohort/task cells. EEGMAT confirmed direction rather than magnitude. Direction was preserved under montage harmonisation and both alternative reference schemes (0/16 sign changes under REST), with near-threshold per-metric attenuations.
Conclusions:
The sensor-level EEG-DMN-like pattern is best interpreted, within the limits of the present evidence, as a proxy of the rest-to-internally-oriented-cognition state transition rather than as a unitary biomarker of default-mode network activity. Reproducibility is component-, window- and estimator-specific: the spectral signature is directionally robust across cohorts, references and montage densities, whereas connectivity depends critically on estimator family and should be reported as such.
