Forewarning of epileptic events from scalp EEG

Abstract

This paper addresses epileptic event forewarning. One novel contribution is the use of graph theoretic measures to detect condition change from time-delay-embedding states. Another novel contribution is better forewarning of the epileptic events from two channels of scalp EEG, with a total true rate of 58/60 (sensitivity = 39/40, specificity = 19/20). Challenges include statistical validation in terms of true positives and true negatives; actionable forewarning in terms of time before the event; detection of the event to reset the forewarning algorithm; and implementation in a practical device.