Identifying an increased risk of epileptic seizures using a multi-feature EEG-ECG classification

Epilepsy, a neurological disorder in which patients suffer from recurring seizures, affects approximately 1% of the world population. In spite of available drug and surgical treatment options, more than 25% of individuals with epilepsy have seizures that are uncontrollable. For these patients with intractable epilepsy, the unpredictability of seizure occurrence underlies an enhanced risk of sudden unexpected death or morbidity. A system that could warn the patient of the impending event or trigger an antiepileptic device would dramatically increase the quality of life for those patients. Here, we proposed a patient-specific algorithm for possible seizure warning using machine learning classification of 34 algorithmic features derived from EEG-ECG recordings. We evaluated our algorithm on unselected and continuous recordings of 12 patients (total of 108 seizures and 3178-h). Good out-of-sample performances were observed around 25% of the patients with an average preictal period around 30 min and independently of the EEG type (scalp or intracranial). Inspection of the most discriminative EEG-ECG features revealed that good classification rates reflected specific physiological precursors, particularly related to certain stages of sleep. From these observations, we conclude that our algorithmic strategy enables a quantitative way to identify "pro-ictal" states with a high risk of seizure generation.