A Shallow Autoencoder Framework for Epileptic Seizure Detection in EEG Signals.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Epileptic seizure detection remains a challenge, particularly for continuous, long-term monitoring.
• Current methods often require complex setups or multiple electroencephalogram (EEG) channels, limiting wearable applications.
• Developing efficient algorithms for single-channel EEG analysis is crucial for accessible patient monitoring.

Purpose of the Study:

• To present a trainable hybrid approach for epileptic seizure detection using a shallow autoencoder (AE) and a conventional classifier.
• To enable at-home diagnosis and monitoring of epileptic patients through wearable devices with minimal EEG channels.
• To evaluate the performance of the proposed method on established EEG datasets.

Main Methods:

• A shallow autoencoder (AE) was trained to minimize signal reconstruction error, generating a low-dimensional feature vector from EEG epochs.
• The encoded AE representation was used as input for k-nearest neighbor (kNN) and support-vector machine (SVM) classifiers.
• The hybrid AE-classifier approach was evaluated on the CHB-MIT and University of Bonn EEG datasets.

Main Results:

• The hybrid AE-kNN method achieved 98.85% accuracy, 99.29% sensitivity, and 98.86% specificity on the CHB-MIT dataset.
• The hybrid AE-SVM method demonstrated high performance with 99.19% accuracy, 96.10% sensitivity, and 99.19% specificity.
• The approach proved effective for single-channel EEG analysis with a 1-second epoch granularity.

Conclusions:

• The proposed shallow AE-based feature extraction offers an effective low-dimensionality representation for epileptic seizure detection.
• The hybrid approach demonstrates superior performance compared to existing methods, suitable for wearable and body sensor networks.
• This method facilitates convenient and continuous epilepsy monitoring in home environments.