Time-Frequency Statistical Features of Delta Band for Detection of Epileptic Seizures

Various research groups are working on the automated detection of epileptic seizures using Electroencephalogram (EEG) data. EEG waveforms are composed of distinct bands of frequencies. Most of the researchers have used a wide range of frequencies or every frequency band of EEG for detection process of epileptic seizures to obtain high accuracy. However, not all frequency bins contain relevant information about seizures, thereby degrading the performance of the detection system. This paper demonstrates the suitability of only delta band (0.5-4 Hz) for the detection of seizures due to epilepsy. The work has been performed in four stages: (1) Short-time Fourier transform (STFT) of EEG data, (2) extraction of delta band from the time-frequency (t-f) plane, (3) calculation of four statistical features (4) performance analysis using Random Forest (RF) classifier. The proposed methodology achieved an average accuracy, specificity and sensitivity of 99.6%, 99.5% and 99.67% respectively between persons suffering from epilepsy and healthy people on Bonn EEG dataset. Proposed work is computationally efficient as it uses only single band which results in small data computation. Its detection time is very short (< 0.5 s) which makes it suitable for real-time clinical application.