Detection and evolution of rhytmic components in ictal EEG using short segment spectra and discriminant analysis

Abstract

An automated method for analysis of ictal EEG is described which aims to reliably detect one or several rhythmic components in short EEG segments (2 sec) and to display their presence, frequency, amplitude, location, and temporal evolution. Spectra were estimated and compared using fast Fourier transform (FFT) and autoregressive modelling (AR). A subsequent linear discriminant analysis decided whether a spectral peak is likely to be caused by rhythmic activity or by the inherent statistical variability. FFT was found to perform better than AR in the detection of rhythmic components, yielding a false-positive rate of 0.825%, a false-negative rate of 2% (signal to noise ratio −4.6 dB), a frequency resolution of 2 Hz, and a temporal resolution of 0.5 sec. Seizure analysis revealed that the ictal scalp EEG of even short seizures can show a complex evolution of rhythmic patterns which are impossible or difficult to recognize by visual inspection or conventional spectral analysis. The following phenomena are demonstrated: superposition of two rhythmic components suggesting two cerebral regions discharging simultaneously and independently with their own pacemakers, sudden and gradual change of frequencies, and gradual development of harmonic frequencies. It is suggested that a more precise correlation between rhythmic generators and seizure symptomatology might allow more predictable pharmacological responses in antiepileptic therapy.