A Complete Strategy for Patient Un-specific Detection of Epileptic Seizures Using Crude Estimations of Entropy

摘要

This paper outlines a complete strategy for patient un-specific detection of epileptic seizures on scalp data. Using a crude estimation of entropy and contextual information derived from methods employed by human experts, a true positive classification rate of 94% was achieved on 125 seizures, 22 different patients and over 500 hours of EEG recordings. False positives remain low enough for this algorithm to be clinically applicable. This paper outlines the strategy, providing justification and exploration on the estimation of entropy using low number of data samples.