Abstracts

Rationale: Epilepsy surgery or RNS may not be a therapy option when seizures are non-localizable or the onset involves widespread network. Open loop thalamic DBS is indicated in this cohort, but frequent stimulations can potentially disrupt physiological activities like sleep and memory. Feedback stimulation in response to an incipient seizure could be advantageous by minimizing stimulation-related side effects. However the challenge is in detecting cortical onset seizures in the thalamus. We aimed to identify biosignatures of the thalamic ictal state using spectral and temporal features using local field potentials (LFP) recorded from the anterior thalamic nuclei (ANT) in 14 consenting adults undergoing stereo EEG investigation for localizing seizures. Methods: Through an IRB approved process, we archived continuous 7-10 days SEEG recordings (including ANT) from 14 subjects (94 seizures). We applied Linelength(LL)1 on seizure onset zone, and ANT to detect seizure state. RandomForest(RF) model2 was trained using features: DWT RWE (9), MSE (29 scales), PSD (9), Hjorth (3), Kurtosis, Skewness, Teager Energy, Linelength, difference between Teager Energy and Linelength and Katz fractal dimension calculated in 4s (3s overlap) using LL detection time +10s seconds with Atlman's modification to select significant features, and testing on all 94 seizures. Results: Features that detected thalamic ictal state were: DWT RWE, the first 2 scales of MSE, PSD 4-8, 13-500, Hjorth, Teager Energy, Linelength, Teager-LL difference and KatzFD. RF outperformed LL by having higher detection rate of thalamic ictal state. Conclusions: Using supervised machine learning, we have discovered candidate biomarkers for thalamic ictal state that hold promise to develop closed-loop DBS in non-localized epilepsies.References1. Pizarro D, Ilyas A, Toth E, et al. Automated detection of mesial temporal and temporoperisylvian seizures in the anterior thalamic nucleus. Epilepsy Res 2018;146:17-20.2. Donos C, Maliia MD, Dumpelmann M, et al. Seizure onset predicts its type. Epilepsia 2018;59:650-  Funding: No funding