Robust seizure detection in EEG using 2D DWT of time-frequency distributions

Seizure is a psychological disorder which is observed through electro-encephalography (EEG) by monitoring the electrical activity inside our brain. Joint time-frequency (TF) analysis is helpful for capturing the dynamic property of EEG signals as solely time or frequency analysis is not sufficient to clearly depict the non-stationary electrical activity. This study is based on the use of 2D discrete wavelet transform (DWT) for feature extraction from the TF image with seizure classification performed through support vector machine. For classifier evaluation a dataset of 80 seizure and 80 non-seizure samples are used, with leave-one-out cross-validation method. The proposed scheme achieved an impressive accuracy of 99.375% and has outperformed state of the art by 6%. This work will assist physicians in seizure detection, particularly in neonatal case where it can cause a life-time disability.