Wavelet analysis as a potential tool for seizure detection

Long-term monitoring of electroencephalograms (EEGs) is a well established clinical procedure for sleep studies and in epileptology for capturing seizure activity and localizing epileptic foci. Good seizure detection schemes facilitate diagnosis of epilepsy and enhance the management of long-term EEG records. The authors investigate a new automated method based on the discrete wavelet decomposition technique for detecting ictal events independent of their EEG morphology. The average power spectrum of EEG of a resting, healthy subject follows a power law attenuation over a range of clinically relevant frequencies suggesting an underlying scale-invariant property. A multi-scale analysis tool is used to monitor this statistical scale-invariance and detect the onset of seizure marked by an appreciable change in this scaling behavior. The detection scheme is based on monitoring the variance structure of the wavelet coefficients over a selected scale range and power fluctuations in these scales individually. Thus, the authors provide a new criteria for the development of an efficient seizure detection scheme