Abstracts

Rationale: Despite advances in medication, surgery, and other therapeutic approaches, many people with epilepsy continue to experience seizures. One of the most difficult features of epilepsy is the unpredictability of seizures, and a reliable warning system for seizures could help patients better manage their lives. Seizure forecasting has been successfully demonstrated using intracranially recorded EEG, but invasive implanted devices may be unsuitable for many patients. Noninvasive wearable biosensors may be capable of forecasting the probability of seizures. However, rigorous testing with EEG-based seizure records is needed to develop and validate such a system. Methods: This three-year project is organized into three phases. In year 1, commercially available wearable sensors are evaluated for signal quality, patient acceptability, and potential to detect and predict seizures in patients undergoing invasive EEG, ambulatory scalp EEG, or hospital scalp EEG. Biosignals are correlated with confirmed clinical and electrographic seizure events. In year two, patients trialing sub-scalp EEG monitoring devices (UNEEG SubQ, or Epi-Minder) and/or ambulatory intracranial EEG device (Medtronic RC+S) will wear sensors for multiple months to correlate biosignal records with EEG seizure annotations. In year three, a machine learning competition to develop forecasting algorithms on biosignals will be conducted. Biosignals from mature, commercially available wearable sensors including photoplethysmography (PPG), accelerometry (ACC), electrodermal activity (EDA), electromyography (EMG), scalp EEG, heart rate (HR), and temperature are under evaluation in year 1. Commercially available biosensors will be evaluated in year 1 for patient acceptance and signal quality, and sensors under evaluation include the Empatica E4 watch, the GeneActiv actigraphy watch, the EpiTel EpiLog scalp EEG sensor, ByteFlies sensor dots, the Biovotion Everion armband, and the Equivital TnR vest. In addition, surveys of mood and premonitory symptoms will be evaluated as possible seizure predictors in years 1 and 2 of the project. Results: To date, we have enrolled 59 patients and recorded 153 seizures over a total of 305 days of monitoring. The enrolled patient group is 53% female with a median age of 31 years. Nineteen patients (32%) were undergoing stereotactic EEG, 29 (49%) ambulatory scalp EEG, 1 (2%) subdural invasive EEG, 9 (15%) in-hospital scalp EEG, an 1 (2%) an implanted device. Eight patients enrolled (14%) were pediatric. Subjects’ seizure localizations were wide ranging, including left temporal (4, 7%), right temporal (3, 5%), right frontal (1, 2%), right occipital (2, 4%), and generalized or non-localized (8, 15%). Twenty seven patients (46%) did not have seizures during monitoring, but may have had behavioral spells or other events. In total 32 seizures have been recorded with EMG, 9 with wireless scalp EEG, 96 with wrist PPG, 37 with chest PPG, 122 with wrist ACC, 76 with chest ACC, and 96 with wrist EDA. Conclusions: The My Seizure Gauge project has made significant progress gathering seizure data using multiple different wearable devices. Funding: Epilepsy Foundation of America - Epilepsy Innovation Institute