Classification of epileptic seizure using hybrid deep learning framework with time and time-frequency Hjorth features

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Signal Processing

Background:

• Epileptic seizures pose significant diagnostic challenges.
• Accurate classification of seizure stages is crucial for effective treatment.
• Existing methods often struggle with complex EEG data patterns.

Purpose of the Study:

• To propose a novel framework for epileptic seizure stage classification.
• To compare time-domain, time-frequency domain, and multi-domain Hjorth parameters.
• To integrate attention-enhanced temporal modeling with deep learning for improved classification.

Main Methods:

• Extraction of Hjorth parameters from time and time-frequency domains (using Discrete Wavelet Transform).
• Implementation of 1D Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) with attention mechanisms.
• Evaluation on the Bonn EEG dataset using 10-fold cross-validation and performance metrics (accuracy, precision, recall, F1-score).

Main Results:

• The multi-domain approach demonstrated superior performance compared to time and time-frequency domain methods.
• Achieved test classification accuracies of 98.40% (binary), 98.00% (three-class), and 85.40% (five-class).
• The proposed framework effectively classifies normal, inter-ictal, and ictal epileptic seizure stages.

Conclusions:

• The proposed multi-domain framework with attention-enhanced temporal modeling offers a robust solution for epileptic seizure classification.
• Hjorth parameters combined with deep learning models show significant potential in analyzing EEG data.
• This framework provides a promising tool for advancing the diagnosis and management of epilepsy.