Abstract
Objective Epilepsy in children is often accompanied by cognitive impairment (CI), causing significant quality of life effects for child and family. Early identification of subjects likely to develop CI could help inform strategies for clinicians. This paper proposes identifying characteristics correlated to CI in preschool children based on electroencephalogram (EEG) network analysis.
Methods A multi-part processing chain analyzed networks from routinely acquired EEG of n = 51 children with early-onset epilepsy (0-5 y.o). Combinations of connectivity metrics (e.g. phase-slope index (PSI)) with network filtering techniques (e.g. cluster-span threshold (CST)) identified significant correlations between network properties and intelligence z-scores (Kendall’s τ, p < 0.05). Predictive properties were investigated via 5-fold cross-validated classification for normal, mild/moderate and severe impairment classes.
Results Phase-dependant connectivity metrics demonstrated higher sensitivity to measures associated with CI, while wider frequencies were present in CST filtering. Classification using CST was approximately 70.5% accurate, improving random classification by 55% and reducing classification penalties by half compared to naive classification.
Conclusions Cognitive impairment in epileptic preschool children can be revealed and predicted by EEG network analysis.
Significance This study outlines identifying markers for predicting CI in preschool children based on EEG network properties, and illustrates its potential for clinical application.
Highlights
EEG network analysis correlates with cognitive impairment in preschool children with epilepsy.
Network sensitivity to impairment improves with dense networks and phase-based connectivity measures.
Classification reveals network features’ predictive potential for clinical impairment identification.