Automated EEG Epileptic Seizure Detection Exploiting Continuous Morlet Wavelet Transform Scalograms and Hybrid Deep Learning Models

Epilepsy is a prevalent neurological disorder characterized by recurrent seizures, affecting millions of people worldwide. Accurate and timely detection of epileptic seizures is crucial for patient management and treatment. Traditional seizure detection methods are often time-consuming and require expert interpretation, leading to a growing interest in deep-learning techniques for efficient and accurate seizure detection. This study ventures into transforming an Electroencephalogram (EEG) signal dataset into Continuous Morlet Wavelet Transform (CMWT) scalograms of the signals processed into four distinct resolutions: 32x32, 48x48, 64x64, and 128x128. CMWT is a mathematical representation that offers a detailed time-frequency-based visualization of signals by decomposing the EEG signals into different frequency components and then visualizing their intensity with respect to time. Five hybrid models are developed and compared: a 3-layered 2D-Convolutional Neural Network (CNN) model, a transfer learning approach with the VGG16 and ResNet50 model, and two hybrid Convolutional Neural Network-Recurrent Neural Network (CNN-RNN) models such as Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) and CNN-LSTM-LSTM. These hybrid models are designed to leverage the strengths of both CNNs and RNNs for feature extraction and classification, respectively. The outcomes demonstrate that the VGG16 architectures with the input resolution 64x64, appear to be a promising choice, offering a harmonious blend of computational efficiency and model performance.