Estimating the time-varying periodicity of epileptiform discharges in the electroencephalogram

Abstract

Periodic lateralized epileptiform discharges (PLEDs) are EEG waveforms that can occur after brain injury or disease. The time-varying periodicity, or instantaneous frequency, of the PLEDs is a potentially important prognostic feature. Estimating the instantaneous frequency, however, is difficult because of the concurrent presence of background activity and artefacts. Here we present a method to enhance the instantaneous frequency features in the joint time-frequency domain. The procedure 1) enhances the PLED spikes in the time-domain using a simple energy operator; 2) transforms to the time-frequency domain using a separable-kernel distribution; and 3) uses a homomorphic filtering approach, within the time-frequency domain, to remove spectral modulation. Existing methods for instantaneous-frequency estimation are then applied to this enhanced time-frequency distribution. We show working examples with EEG epochs but have yet to test the method over an entire EEG database.