Abstracts

Rationale: Epilepsy is a neurological disorder characterized by recurring seizures. It affects around 3 million people in the US and ~1% of the global population. Around 30-40% of those affected cannot achieve an acceptable level of control of their seizures via drug treatment. Epilepsy surgery and implantable stimulation devices are options that can reduce seizures. An essential part of these treatments is the information of the SOZ. Ictal localization (the current clinical standard) requires 7-14 days of costly hospitalization with accompanying patient discomfort and infection risk. Experimental methods for interictal localization predominantly use a single biomarker; e.g., High Frequency Oscillations (HFOs), Interictal Epileptiform Discharges (IEDs), Phase Amplitude Coupling (PAC), etc. These methods do not scale due to patient and disease specific characteristics as individual biomarkers may not be observed in all cases. Hence, we developed a method to identify patient-specific interictal biomarkers for localization of SOZ. Methods: Analysis on 2-hour interictal iEEG segments of 9 patients with temporal lobe focal epilepsy was performed. Data was sampled at 5,000 Hz with a 1000 Hz anti-aliasing filter. Pre-ictal and discontinuous recordings were excluded while selecting the data segments. Three widely used biomarkers (HFO[Matsumuto 2013], IED[Barkmeier 2012], and PAC[Weiss 2015]) were extracted in the 2-hour segments. In addition, pathological abnormal events (HFOs, IEDs and PAC) in the first minute of the analysis segments were manually annotated by visualizing the EEG. Biomarkers extracted in the first minute were correlated with the annotations to select a biomarker that had the highest correlation. Detections of that biomarker in the 2-hour segments were then utilized in an already published framework that deploys clustering and Bayesian filtering to assign likelihood probabilities for the channels for being in SOZ [Varatharajah 2017]. Results: Resulting likelihood probabilities were compared against the gold standard SOZ channels (determined by an epileptologist) and ROC analysis was utilized to assess the proposed approach. Our approach results in an average Area Under ROC Curve (AUC) of 0.81 for 9 patients. Utilizing the same biomarker for all the patients instead of the pre-selected biomarker results in lower AUCs; i.e., PAC: 0.79, HFO: 0.76, and IED: 0.73. In addition, utilizing clustering and Bayesian filtering improves the localization accuracies significantly in all three biomarkers (p < 0.05, paired t-test). Conclusions: This study confirms that individualization of the biomarker utilized in interictal SOZ localization is feasible. We developed and validated a lightly-supervised technique which can pre-select one among several biomarkers. Results indicate that this method outperforms approaches which utilize only one of the biomarkers for every patient. However, studies involving larger number of patients, and including different sleep states are necessary to ensure robustness of this approach. Funding: Mayo Clinic and Illinois Alliance Fellowship, NIH grants NINDS-R01-NS92882, NHLBI-HL105355, and NINDS-UH2-NS095495-01, NSF grants CNS-1337732 and CNS-1624790, FNUSA-ICRC: Project no. LQ1605 from the National Program of Sustainability II (MEYS CR), and São Paulo Research Foundation 2014/01587-8.