Seizure detection in EEG signal using Gaussian-stockwell transform and Hermite polynomial features

Abstract

Electroencephalography (EEG) is a non-intrusive and powerful tool for understanding the complex nature of the brain. Epileptic seizures, a neurological disorder, are characterized by excessively synchronized neuronal activity within the brain. This abnormal synchronization disrupts the normal functioning of neural networks, leading to seizures. This work proposes to use the Gaussian-based Stockwell Transform (GST) to segment EEG signals into various frequency bands such as delta, theta, alpha, beta, and gamma band for analysis. The feature extraction process on these sub-band signals generates a feature set, encompassing the fluctuation index, relative amplitude, holoentropy, logarithmic band power, and the newly introduced Hermite polynomial-based delta amplitude modulation spectrogram. To differentiate between epileptic and non-epileptic events in EEG signals, the previous features have been applied to the Random Forest Classifier (RFC) algorithm. The effectiveness of our proposed method is validated using a benchmark EEG dataset, with GST and RFC demonstrating accurate results in distinguishing seizure activity. Experimental results show an optimal classification accuracy of 96.4 %, a sensitivity of 97 %, and a specificity of 96.4 %.