Wendy Xiong, Tatiana Kameneva, Elisabeth Lambert, Ewan Nurse
Epilepsy is a neurological disorder characterized by recurrent seizures that are transient symptoms of synchronous neuronal activity in the brain. Epilepsy affects more than 50 million people worldwide [1]. In Australia, over 225,000 people live with epilepsy [2]. Seizure prediction allows patients and caregivers to deliver early interventions and prevent serious injuries. Electroencephalography (EEG) has been used to predict seizure onset, with varying success between participants [3, 4]. There is an increasing interest to use electrocardiogram (ECG) to help with seizures detection and prediction. The aim of this study is to use long-term continuous recordings of EEG and ECG data to forecast seizures.
EEG and ECG data from 7 patients was used for analysis. Data was recorded using 21 EEG electrodes and 3 ECG electrodes by Seer with an ambulatory video- EEG-ECG system. The average period of recording was 95 hours (range 51-160 hours). Data was annotated by a clinician to indicate seizure onset and offset. On average, 4 clinical seizures occurred per participant (range 2-10). EEG and ECG data were bandpass filtered using Butterworth filter (1-30 Hz for EEG, 3-45 Hz for ECG).
A characteristic of a system that is nearing a critical transition is critical slowing, which refers to the tendency of the system to take longer to return to equilibrium after perturbations, measured by an increase in signal variance and autocorrelation [5]. The variance and autocorrelation of EEG and ECG signals were calculated for each electrode in 1 s window for each time point. The autocorrelation value was set to the width of half maximum of the autocorrelation function. The instantaneous phases of variance and autocorrelation signals were calculated at each time point using Hilbert transform. To extract long (1 day) and short (20 s in EEG, 10 min in ECG) cycles in the variance and autocorrelation signals, a moving average filter has been applied. The relationship between seizure onset times and phase of variances and autocorrelation were investigated in long and short cycles. The probability distribution for seizure occurrence was determined for each time point. The seizure likelihood was determined at three levels: low, medium and high, based on two thresholds defined as functions of maximum seizure probability. Data analysis was performed in Python 3.
Results show that the variance and autocorrelation of EEG data increased at the time of seizure onset in 66.7% and 68.3% of cases, respectively. The variance and autocorrelation of ECG data increased at the time of seizure onset in 60% and 50% cases, respectively. Long and short cycles of variance and autocorrelation had consistent results. Result indicate that critical slowing may be present in a neural system during seizures and this feature could be used to forecast seizures.
References
[1] Thijs et al. Epilepsy in adults The Lancet, 2019
[2] Facts and Statistics, Epilepsy Action Australia, www.epilepsy.org.au
[3] Cook et al Prediction of seizure likelihood with a long-term, implanted seizure advisory system in patients with drug-resistant epilepsy: a first-in- man study. The Lancet Neurology, 2013
[4] Karoly et al The circadian profile of epilepsy improves seizure forecasting. Brain, 2017
[5] Scheffer et al Early-warning signals for critical transitions. Nature, 2009
