Robust multi-centre interictal EEG biomarker for distinguishing epilepsy from mimickers

Area of Science:

• Neuroscience
• Clinical Neurology
• Biomedical Engineering

Background:

• Epilepsy diagnosis relies heavily on electroencephalography (EEG), but distinguishing it from mimic conditions is challenging.
• Prior EEG biomarkers often suffer from limited generalizability due to single-center validation and small sample sizes.
• There is a critical need for robust, interpretable, and multi-center validated EEG biomarkers for epilepsy diagnosis.

Purpose of the Study:

• To develop and validate an interpretable, multi-center EEG biomarker for differentiating epilepsy from mimic conditions.
• To address the limitations of previous single-center studies with small sample sizes.
• To establish a reliable decision-support tool for epilepsy diagnostics.

Main Methods:

• Analysis of routine interictal EEG data from 448 subjects across two tertiary centers (IHBAS and MAX).
• Computation of a 13-dimensional Interictal Clinical Signature (ICS) based on spectral slowing, posterior dominant rhythm, complexity, and network synchrony.
• Application of a two-stage logistic regression framework for subject-level predictions and assessment of recording-length confounding factors.

Main Results:

• Achieved within-center Area Under the Curve (AUC) values of 0.723 (IHBAS) and 0.790 (MAX).
• Demonstrated symmetric cross-center generalization (AUC 0.725) when recording length was excluded as a predictor.
• Confirmed consistent cross-center performance (AUC 0.70-0.72) with fixed-duration truncation and stability across sampling rates (20-125 Hz).

Conclusions:

• Low-dimensional, interpretable EEG features, encapsulated in the ICS, enable robust epilepsy versus mimic classification.
• The multi-center validation confirms the ICS framework's utility as an accessible, center-invariant decision-support tool for epilepsy diagnostics.
• The systematic analysis of confounding factors provides a methodological template for future EEG biomarker research.