Deep learning approach to detect seizure using reconstructed phase space images.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Epilepsy is a chronic neurological disorder affecting brain function, detectable via electroencephalogram (EEG) signals.
• EEG signals exhibit nonlinear and chaotic dynamics, necessitating advanced analysis techniques.
• Previous epilepsy detection methods often simplify seizure classification, overlooking nuanced states.

Purpose of the Study:

• To develop an accurate epilepsy detection system using reconstructed phase space (RPS) images of EEG signals.
• To apply transfer learning with deep neural networks for enhanced seizure classification.
• To evaluate the efficacy of Convolutional Neural Networks (CNNs) in distinguishing between normal, interictal, and ictal EEG states.

Main Methods:

• EEG signals were transformed into reconstructed phase space (RPS) images.
• A pre-trained deep neural network model was utilized and retrained with RPS images via transfer learning.
• The CNN model's performance was evaluated for both binary (normal vs. ictal) and ternary (normal vs. interictal vs. ictal) classification.

Main Results:

• The CNN model achieved high classification accuracy: (98.5±1.5)% for binary and (95±2)% for ternary classification.
• The proposed CNN approach outperformed existing statistical methods in accuracy, sensitivity, and specificity.
• RPS images combined with CNN demonstrated significant potential for epileptic seizure prediction.

Conclusions:

• Employing RPS images with CNNs offers a promising approach for accurate epileptic seizure detection.
• The study highlights the effectiveness of capturing nonlinear EEG dynamics for improved neurological disorder analysis.
• This method provides a robust framework for advancing the diagnosis and management of epilepsy.