Beyond Regional Biomarkers: Multi-Axis Spectral-Dynamical Architecture and Normative Deviation Panels for Resting EEG Across Psychiatric and Neurodevelopmental Disorders

Psychiatric and neurodevelopmental disorders are commonly studied using localized EEG biomarkers such as band-power differences or region-specific abnormalities. While such approaches have produced valuable insights, they often show limited consistency across datasets and diagnostic groups, suggesting that disorder-relevant signal may reside in the large-scale organization and regulation of neural activity rather than in isolated features.

This work introduces a multi-axis spectral-dynamical architecture framework for analyzing resting-state EEG. The framework organizes EEG-derived features into three conceptual layers:

• Layer 1: Window-level spectral-dynamical descriptors capturing entropy-like structure, spectral peak organization, and temporal variation

• Layer 2: Cortical architecture features summarizing the spatial distribution of these measures across anatomically interpretable axes (e.g., frontal–posterior, left–right)

• Layer 3: Higher-order regulation features capturing temporal instability, condition-response (eyes-open vs eyes-closed), cross-feature coupling, composite indices, and multivariate deviation from healthy reference structure

Across multiple public EEG datasets spanning healthy and clinical populations, we observe that resting EEG is dominated by a frontal–posterior architectural organization, but diagnostic differences are expressed more strongly in variance structure, temporal instability, coupling relationships, and condition-response behavior than in mean feature shifts alone.

A multifeature panel-based approach yields a Normative Deviation Index (NDI) that quantifies deviation from healthy resting EEG organization and demonstrates strong discrimination between healthy and clinical populations (ROC-AUC > 0.82 in primary contrasts). In parallel, an architecture-oriented panel preserves interpretability of cortical organization and feature-family contributions.

These results support a shift from isolated EEG biomarkers toward panel-based, multi-axis characterization of resting brain organization, in which psychiatric and neurodevelopmental conditions are understood as alterations in the structure, regulation, and stability of cortical dynamics.

Notes

• This Zenodo release provides a conceptual and empirical overview of the framework along with figures and summary results.
• Detailed implementation logic, optimization procedures, and full feature construction pipelines are not included in this version.
• Supporting materials may be shared for academic discussion upon request.