Classification of Epileptic Seizures using Optimized TQWT and Hybrid Models

Abstract

Epilepsy is a Central Nervous System (CNS) disorder that can cause chronic seizures at any time. The electroencephalogram (EEG) records the electrical activities caused by the postsynaptic potentials that can be used to diagnose any disorder in the brain. This study identifies the best methods of forecasting epileptic seizures by comparing different approaches. The EEG usually contains enormous data, which becomes time-consuming and laborious for data interpretation. This study proposes to develop a learner with an automated signal interpretation technique using advanced signal processing methods that can predict seizures from the EEG recordings. The extracted EEG signals from the patient are subjected to an optimized tunable Q-factor wavelet transformation. The global, temporal, and entropy-based features are extracted from the sub-bands and fused. An ANN model is trained with the fused features. Also, from the TQWT subbands, EEG scalograms are generated and used to train a CNN model. These models are trained in such a way that they can differentiate between the normal, ictal, and interictal classes. The performance of the CNN model trained with scalogram images by the proposed approach is compared to the performance of deep hybrid models. The ANN hybrid model produced an accuracy of 98% using different categories of features extracted, and the CNN model produced an accuracy of 91 % using scalogram images of EEG signals, which outperformed the hybrid model in terms of speed and computation.