Classifying Single Channel Epileptic EEG data based on Sparse Representation using Shallow Autoencoder.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Machine Learning

Background:

• Epileptic seizure detection from electroencephalogram (EEG) data is crucial for patient care and management.
• Current wearable health systems require computationally efficient and accurate seizure detection algorithms.
• Patient-independent algorithms are essential for broad applicability in real-world scenarios.

Purpose of the Study:

• To develop a patient-independent epileptic seizure detection algorithm for scalp EEG.
• To integrate neural and conventional machine learning for improved wearable health systems.
• To enhance accuracy and reduce computational complexity in seizure detection.

Main Methods:

• A shallow autoencoder model was used for sparse representation of single-channel EEG signals.
• A k-nearest neighbor (kNN) classifier was employed to categorize EEG data as epileptic or non-epileptic.
• The optimal sparsity level for the autoencoder was investigated, with a level of 4 yielding the best results.

Main Results:

• The proposed method achieved high performance metrics: 98.85% accuracy, 99.29% sensitivity, and 98.86% specificity on the CHB-MIT scalp EEG database.
• The algorithm outperforms existing state-of-the-art seizure detection methodologies.
• Shallow learning approach proved computationally lighter than deep learning for feature extraction.

Conclusions:

• The developed algorithm offers a computationally efficient and highly accurate solution for patient-independent epileptic seizure detection.
• The integration of shallow autoencoders and kNN classifiers demonstrates a promising approach for advancing wearable epilepsy monitoring.
• The method's suitability for single-channel EEG processing makes it ideal for integration into small, wireless sensing devices.