Comparison study of seizure detection using stationary and nonstationary methods

Abstract

We present an accurate seizure detection algorithm, and make a detailed comparison of two frequency analysis methods: a widely used stationary method - Fast Fourier Transform (FFT) and a relatively new nonstationary method - Hilbert-Huang Transform (HHT). Two public databases and one our own database were tested. The results show that our algorithm has very high accuracy compared with the state-of-the-art. More interestingly, it shows that the nonstationary method HHT offers better performance than the stationary method FFT in seizure detection. Therefore we propose that we should pay attention to the nonstationarity of EEG signal, since the "stationary assumption" may introduce some inaccuracy.