Multiple combinations of time-frequency distributions and classifiers for detecting epileptic EEG signal

Inhalt:
Epileptic seizure is a sudden, abnormal discharge of neurons in the brain that leads to a chronic disease of transient brain dysfunction. Electroencephalogram (EEG) have an irreplaceable role in the diagnosis of epilepsy, because it helps doctors classify the epileptic syndrome and locate epileptic lesions. The most universal method for processing such dynamic and nonstationary signal is wavelet transform (WT), and the previous research has paid little attention to the difference between time-frequency (t-f) distributions, for example short-time Fourier transform (STFT) and Cohen’s class. In this article, we compare different kinds of time-frequency distributions and classifiers. Totally 14 kinds of t-f distributions and 23 classifiers are included to make up 322 kinds of combination. The best combination of WT and quadratic discriminant analysis (QDA) shows its great performance with 100% accuracy in two classification task, 99.7% in three classification task and 99.3% in five classification task.