Minimum data requirements and automated preprocessing for reliable EEG biomarkers in Rett syndrome

Electroencephalography (EEG) is a promising biomarker for Rett syndrome (RTT), but excessive artifact and variable tolerance for longer recording sessions pose challenges for reliable biomarker development. Establishing an automated preprocessing pipeline that matches human review and defining the minimum data needed for stable quantitative EEG (qEEG) features can support more patient-friendly protocols and provide consistent multisite analysis results.

A mean of 10 min of resting-state EEG from 117 participants (1-18 year old; 236 sessions) in the multisite R61 RTT study was processed using a fully automated, correction-based preprocessing pipeline incorporating artifact handling, adaptive channel rejection, ASR, and ICA-based cleaning. Spectral power was extracted from artifact-free 4-s epochs. The proposed pipeline is validated using an established rejection-based pipeline. Feature stability as a function of cumulative data length was then assessed using two complementary frameworks: a Statistical Convergence approach and a Model-Based Inflection approach, and potential systemic dependencies were evaluated using permutation analyses. The relationship between clinical measures was also assessed.

The correction-based pipeline retained substantially more data than the rejection-based workflow (mean retention = 95.0% vs. 28.4%; p < 0.001) while preserving strong feature correspondence across frequency bands. Stable power estimates were achieved after 19-34 epochs (= 76-136 s). Based on permutation analysis, there was no statistically significant difference in minimum stabilization threshold between RTT and TD. However, the RTT group exhibited higher rates of intrinsic signal instability than typically developing (TD) controls. Age-stratified analysis revealed that the minimum epochs did not significantly differ between age groups. Spectral associations with clinical severity were preserved when using only the minimum data required for stability, as well as in an ecologically valid scenario of truncating the raw EEG up to minimum epoch recommendation and reprocessing it.

With the proposed correction-based pipeline, approximately 3 min of raw resting-state EEG are sufficient to obtain stable and clinically meaningful spectral features in children with Rett syndrome. These findings support shorter, more feasible EEG acquisitions and provide a reproducible framework for data sufficiency in multisite neurodevelopmental studies.