On-line neonatal seizure detection based on multi-scale analysis of EEG using wavelets as a tool

Seizures represent the most distinctive sign of neurologic disease in the neonate. The definitive method to identify seizures is based on the visual analysis of the electroencephalogram (EEG). Reliable seizure detection in neonates can optimize clinical treatment and theoretically reduce brain injury. Most efforts in computerized, automated seizure detection have been directed towards adults and hence commercially available schemes are not specifically focused on neonatal seizure detection. As in adults, the normal spectrum of neonatal EEG follows an inverse power-law attenuation over a band of clinically relevant frequencies suggestive of self-similar fluctuations over a multiplicity of scales. In this paper, measures to monitor the scaling property of the neonatal EEG to detect electrographic seizures are proposed. This is achieved by multi-scale analysis of the signal using the wavelet transform as a tool. A seizure detection scheme which can be implemented on-line is proposed. The test set included data from five neonates of 36-42 weeks conceptional age. Preliminary tests on 18 channels of data from 5 patients yielded 95.9% seizure detection rate. The detection rate is 100% when the analysis is based on multichannel data