P89. EEG as a biomarker in experimental epileptogenesis

Abstract

Objective Epileptic seizures are characterized by spontaneous and mostly rhythmic discharges of groups of neurons and can be recorded by scalp and intracranial electroencephalography (EEG). Mesial temporal lobe epilepsy with hippocampal sclerosis (mTLE-HS) is a characteristic entity that often develops after an initial precipitating injury (IPI). The pathophysiological changes leading to epilepsy during this time between an IPI and the manifestation of epilepsy is called epileptogenesis. It is clinically silent and as of now cannot be detected. For developing a preventive treatment, a reliable biomarker of epileptogenesis needs to be established. EEG data may provide a suitable basis. The aim of this work was to computationally characterize, quantify and classify intra-hippocampal EEG recordings in a rat mTLE-HS model with respect to the development of epilepsy. Methods mTLE-HS was induced in 4 male Sprague–Dawley rats by repeated electrical stimulation of the perforant pathway ( Norwood et al., 2010 ). After 2–4 weeks, spontaneous seizures of hippocampal origin occurred. Intracranial EEG was continuously and wirelessly recorded unilaterally from the granule cell layer of the dentate gyrus. In total, 20 days of baseline (BL), 81 days of epileptogenesis (EPG) and 162 days of manifest epilepsy (MFE) were recorded and preprocessed for analysis. For feature extraction, open source software ( ECP19V1 ) was used to quantify 1-s-EEG epochs with 7 characteristics: Power, Coastline, Intermittency, Coherence, Asymmetry, Rhythm and Spikiness. Linear mixed-effects modelling ( Maechler et al., 2015 ) was performed for each characteristic to show inter- and intra-individual differences between BL-EPG, EPG-MFE and BL-MFE.A simple three-layered, feed-forward neural net was trained to classify independent EEG epochs as either BL, EPG or MFE ( R-project.org/package=caret ). Results Only Asymmetry showed congruent changes over the time course inter-individually. For all other characteristics, significant changes still could be shown intra-individually. The neural net achieved a positive predictive value of 61% (balanced accuracy: 68%). Conclusion Our single channel EEG data set currently represents 4 whole-time courses ‘from healthy to epileptic’ in rats. The 7 characteristics proved to be useful for EEG feature extraction and data reduction. For further analyses, more features (e.g. power in different frequency bands) will be included. Analyzing independent 1-s EEG epochs, the simple neural net performed surprisingly well. The use of a long short-term memory net is in preparation and should improve the classification performance by considering the time series properties. The applicability and predictive performance of our findings in rats will be tested on human intracranial recordings.