Multi-channel EEG based Neonatal Seizure Detection

A multi-channel method for patient specific and patient independent, EEG based neonatal seizure detection is presented. Two classifier configurations are proposed and tested, along with a number of classifier models. Existing methods for neonatal seizure detection have been empirical threshold based or based on a single EEG channel. The optimum patient specific classifier for EEG based neonatal seizure detection was found to be an Early Integration configuration employing a linear discriminant classifier model. This yielded a mean classification accuracy of 74.66% for 11 neonatal records. The optimum patient independent classifier was an Early Integration configuration with a linear discriminant classifier model giving a mean accuracy of 72.81%