Entropy-driven deep learning framework for epilepsy detection using electro encephalogram signals

Abstract

Epilepsy is one of the most frequently occurring neurological disorders that require early and accurate detection. This paper introduces a novel approach for the automatic identification of epilepsy in EEG signals by incorporating advanced entropy-based measures with modern pre-processing techniques. The objective is to develop a robust and effective epilepsy detection method. EEG data were pre-processed using adaptive wavelet denoising models to suppress noise while preserving signal integrity. Multivariate entropy features, including Multiple Variable Permutation Entropy (mvMPE) and Multiple Variable Multi-Scale Fuzzy Entropy (mvMFE), were extracted to capture both complexity and frequency-specific variations. Additionally, Uniform Manifold Approximation and Projection (UMAP) was applied for non-linear dimensionality reduction, enhancing the discriminative power of features. A Residual Convolutional Neural Network (ResNet) integrated with Bi-Directional Long Short-Term Memory (Bi-LSTM) was employed to capture both temporal dynamics and spatial features. The proposed model demonstrated superior classification accuracy compared to traditional approaches. Implemented using Python, the model achieved an accuracy of 94%, F1-Score of 96%, recall of 93%, specificity of 87.70%, and precision of 82.21%. This study highlights the synergy between advanced entropy measures and cutting-edge deep learning architectures for robust and accurate epilepsy detection.