Classification of Epileptic EEG Signals Using Time-Frequency Analysis and Deep Learning

Diagnosing and tracing of neurological disorders using Electrencephalography (EEG) signals is a promising and popular approach. For analyzing the EEG signals, time-frequency (TF) analysis is effective and useful method because of the stochastic behavior of the signals. Wavelet Transform (WT) was used broadly for analyzing the EEG signals and successful results were reported for various neurological disorders. As a new and promising TF method, concentration of frequency and time (ConceFT) was preferred to use in this study. This new TF method improves the TF resolution obtained using WT. ConceFT combines the multitaper technique and the synchrosqueezing transform (SST). This approach increases the TF resolution and the accuracy of the energy representation. In this paper, an algorithm designed for distinguishing the stages of epileptic seizure using ConceFT was presented. TF images of EEG signals were produced using ConceFT and these images were fed into SqueezeNet. The aim of this approach is to show the success of ConceFT in analyzing the non-stationary signals. Classifications of epileptic EEG signals were performed according to different scenarios. Classification accuracies were obtained between 91,18% and 100%, which showed the promising performance of ConceFT.