Objective: Cognitive impairment (CI) is common in children with epilepsy and can have devastating effects on their quality of life. Early identification of CI is a priority to improve outcomes, but the current gold standard of detection with psychometric assessment is resource intensive and not always available. This paper proposes exploiting network analysis techniques to characterize routine clinical electroencephalography (EEG) to help identify CI in children with early-onset epilepsy (CWEOE) (0-5 years old).
Methods: Functional networks from routinely acquired EEGs of 51 newly diagnosed CWEOE were analyzed. Combinations of connectivity metrics with subnetwork analysis identified significant correlations between network properties and cognition scores via rank correlation analysis (Kendall's τ). Predictive properties were investigated using a cross-validated classification model with healthy cognition, mild/moderate CI, and severe CI classes.
Results: Network analysis revealed phase-dependent connectivity having higher sensitivity to CI and significant functional network changes across EEG frequencies. Nearly 70.5% of CWEOE were aptly classified as having healthy cognition, mild/moderate CI, or severe CI using network features. These features predicted CI classes 55% better than chance and halved misclassification penalties.
Conclusions: Cognitive impairment in CWEOE can be detected with sensitivity at 85% (in identifying mild/moderate or severe CI) and specificity of 84%, by network analysis.
Significance: This study outlines a data-driven methodology for identifying candidate biomarkers of CI in CWEOE from network features. Following additional replication, the proposed method and its use of routinely acquired EEG forms an attractive proposition for supporting clinical assessment of CI.
Keywords: Developmental impairment; EEG graph networks; Network analysis; Pediatric epilepsy; Signal processing.
Copyright © 2018 Elsevier Inc. All rights reserved.