Authors :
Presenting Author: Daniel Lachner-Piza, PhD – 1Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute
Julia Jacobs, Dr. – 1Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute; Philippe Kahane, Dr. – 4Univ. Grenoble Alpes, CHU Grenoble Alpes, Grenoble Institut des Neurosciences; Jan Schönberger, Dr. – 2Department of Neuropediatrics and Muscular Disease, Medical Center–University of Freiburg
Rationale:
Radio frequency thermocoagulation (RFTC) induces very focal lesions and aims to ablate seizure onset areas and reduce the seizure frequency in patients with epilepsy. RFTC can be performed via the same electrodes used to record the intracranial electroencephalogram (EEG). We aim to predict the success of an RFTC intervention by analyzing the induced change in the activity from High Frequency Oscillations (HFO) and their subgroup coinciding with interictal epileptic spikes (iesHFO).
Methods:
Our study included 21 patients. The HFO and iesHFO biomarkers were detected automatically and their activity was characterized by their occurrence rate, amplitude and power. We first analyzed the anatomical and connectome scope of the RFTC induced change on the biomarkers, by defining six different brain zones that varied in their anatomical and electrophysiologic connectivity to the thermocoagulation site: (i) rftcSite; (ii) rftcConnected: zones with a high connectivity to rftcSite; (iii) highEI: zones with a high epileptogenicity index; (iv) rftcStructure: zones within the same anatomical structure as rftcSite; (v) rftcLobe: zones within the same lobe as rftcSite; (vi) rftcHemisphere: zones within the same hemisphere as rftcSite. The electrophysiologic connectivity between zones was calculated using multiband correlation. Each brain zone was characterized by the activity Δ, (i.e. the post-RFTC minus the pre-RFTC biomarker activity). The patients were pooled into a good-outcome group if their seizure frequency reduction was at least 90% (Fig. 1). We tested if the activity Δ from a given brain zone was higher for the good-outcome patients than for the bad-outcome patients (Mann-Whitney U test, p < 0.0005); if the test was passed, the brain-zone specific activity Δ was selected as a relevant feature for the prediction of outcome. The k-means clustering (k=2) algorithm was used for the outcome prediction.