Seizure Pattern Recognition Using Fast Walsh Hadamard (FWHT) Transform

Epileptic seizure detection is an emerging approach to the neurological processing of brain signals. Previously, recognition of epileptic seizures has been done from visual scanning of EEG signals by expert neurologists into various categories such as healthy and fatigue epochs. Unfortunately, this procedure is not effective because it is exhausting, time consuming and generally leads to incorrect results. Therefore, specialists found that automatic seizure detection a more effective technique for diagnosing. In this paper, we demonstrate an automated method for detection of abrupt changes of Electro-Encephalogram (EEG) signals depending on usage of FAST Walsh Hadamard Transform (FWHT). The FWHT analyzes the EEG signals in the frequency domain and decomposes it into the Hadamard coefficients. Different signal attributes are extracted from the decomposed EEG signals. These attributes comprise: band power Beta, band power Delta, band power Gamma, band power Theta and finally mean curve length (MCL) attribute. Finally, classification is implemented using a thresholding strategy to discriminate between seizure and healthy epochs. This method is tested on long-term EEG recordings from the available Physio-Net EEG dataset. The proposed method demonstrates a high classification performance in comparison with other previous methods. An average sensitivity of 94.54%, an average specificity of 95.2% and an average accuracy of 96.1363% are achieved from the mean curve length feature with FWHT. Keywords: EEG, epilepsy, seizure detection and FWHT