Authors :
Presenting Author: Suzanne van Norden, MSc – University of Twente and Medisch Spectrum Twente (MST)
Prejaas Tewarie, MD PhD – University of Twente and Medisch Spectrum Twente (MST); Kuan Kho, MD PhD – University of Twente and Medisch Spectrum Twente (MST); Anne Marthe Meppelink, MD PhD – Stichting Epilepsie Instellingen Nederland (SEIN); Jacqueline Ardesch, MD – Stichting Epilepsie Instellingen Nederland (SEIN); Marleen Tjepkema-Cloostermans, PhD – University of Twente and Medisch Spectrum Twente (MST); Michel van Putten, MD PhD – University of Twente and Medisch Spectrum Twente (MST)
Rationale:
The excitation-inhibition (E/I) balance is generally assumed to be elevated in patients with epilepsy, leading to an increased risk of interictal epileptiform discharges (IEDs) and seizures. A reliable and robust method to estimate this property from EEG recordings may assist in guiding and monitoring treatment. Here, we apply two methods to estimate the E/I ratio from ultra-long subcutaneous continuous EEG and relate it to the occurrence of IEDs.
Methods:
We used data from patients with frontotemporal or generalized epilepsy who were included in our ongoing (PREDYct) study. Patients recorded two-channel EEG with a subcutaneous electrode (UNEEG Medical) positioned over the left or right frontotemporal area. All IEDs for the first two weeks were visually assessed and annotated. We subsequently calculated the E/I ratio for all nights (12am -7am) using two different methods to study its relationship with the rate of IEDs.
We automatically detected all spikes and waves of the IEDs with a custom built algorithm. For the second method, we used a corticothalamic mean-field model (Abeysuriya et al, Journal of Neuroscience Methods, 2015) which estimates the E/I ratio from the EEG power spectrum, expressed as the ratio of the excitatory and inhibitory cortical gain. For every patient, we compared the results of both methods with the annotated IED rate. We used Monte Carlo simulation to assess significance.
Results:
We included four patients. In three patients, IEDs were characterized by spike-wave discharges. Our algorithm performed satisfactory in automated detection of spike amplitude and wave area (Figure 1A). Figure 1B shows an example of a fitted power spectrum by the mean-field model. Figure 2A presents an example of the WA/sa, the E/I ratio and the IED rate. We found the strongest relation between the E/I ratio as assessed with the mean-field model and the IED rate. For all patients, this relation was significant (p < 0.01), with increasing IED rates for higher E/I ratios (Figure 2B).