Automatic diagnosis of epileptic seizures using entropy-based features and multimodel deep learning approaches

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epilepsy is a common neurological disorder characterized by recurrent seizures.
• Seizures pose risks including injury and fatality, necessitating predictive measures.
• Electroencephalogram (EEG) analysis offers a pathway for predicting seizures.

Purpose of the Study:

• To develop an automatic approach for diagnosing epileptic seizures in children using EEG data.
• To compare the efficacy of machine learning (ML) and deep learning (DL) models for seizure prediction.

Main Methods:

• Extracted time-domain features (Hurst exponent, Tsallis entropy, enhanced permutation entropy, amplitude-aware permutation entropy) from EEG data.
• Classified EEG data using ML models (SVM, KNN, DT) in the first session.
• Classified EEG data using DL models (GRU, LSTM, BiLSTM RNN) in the second session.

Main Results:

• Deep learning models, specifically GRU, LSTM, and BiLSTM RNN, demonstrated superior performance.
• The fusion of time-domain and entropy features with RNN classifiers enhanced classification accuracy.
• The study successfully differentiated epileptic seizures from normal EEG patterns in children.

Conclusions:

• Recurrent Neural Network (RNN) deep learning classifiers, particularly GRU, LSTM, and BiLSTM, are effective for automatic epileptic seizure detection.
• Combining time-domain and entropy-based features significantly boosts the performance of DL models in seizure prediction.
• This approach holds promise for developing early warning systems for epilepsy patients.