Epilepsy, characterized by recurrent, unpredictable seizures, is one of the most common neuropsychiatric disorders facing mankind. It is also one of the most debilitating. Apart from the neurodegenerative and dissociative effects of acute seizure, a major factor in the burden inherent to epilepsy is the unpredictability of upcoming seizures, which increases risk of injury and enhances comorbidity of psychosocial disorders. A system that could reliably deliver warnings before seizure would restore a feeling of control to patients, while accelerating the development of new treatment options.
Ten seizures were isolated offline by extracting neural activity from intracranial recordings across two patients diagnosed with intractable epilepsy. Analyses were performed to search for dynamics in brain activity preceding seizure onset. Then, SVMs and CNNs were trained to distinguish clips of data sampled from either before or between seizures. This work describes the results and provides suggestions for future progress in seizure prediction.