A computational biomarker of idiopathic generalized epilepsy from resting state EEG

Abstract

<jats:title>Summary</jats:title><jats:p>Epilepsy is one of the most common serious neurologic conditions. It is characterized by the tendency to have recurrent seizures, which arise against a backdrop of apparently normal brain activity. At present, clinical diagnosis relies on the following: (1) case history, which can be unreliable; (2) observation of transient abnormal activity during electroencephalography (<jats:styled-content style="fixed-case">EEG</jats:styled-content>), which may not be present during clinical evaluation; and (3) if diagnostic uncertainty occurs, undertaking prolonged monitoring in an attempt to observe <jats:styled-content style="fixed-case">EEG</jats:styled-content> abnormalities, which is costly. Herein, we describe the discovery and validation of an epilepsy biomarker based on computational analysis of a short segment of resting‐state (interictal) <jats:styled-content style="fixed-case">EEG</jats:styled-content>. Our method utilizes a computer model of dynamic networks, where the network is inferred from the extent of synchrony between <jats:styled-content style="fixed-case">EEG</jats:styled-content> channels (functional networks) and the normalized power spectrum of the clinical data. We optimize model parameters using a leave‐one‐out classification on a dataset comprising 30 people with idiopathic generalized epilepsy (<jats:styled-content style="fixed-case">IGE</jats:styled-content>) and 38 normal controls. Applying this scheme to all 68 subjects we find 100% specificity at 56.7% sensitivity, and 100% sensitivity at 65.8% specificity. We believe this biomarker could readily provide additional support to the diagnostic process.</jats:p>