A quadratic linear-parabolic model-based EEG classification to detect epileptic seizures

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Neuroscience

Background:

• Epileptic seizures are neurological disorders characterized by abnormal brain activity.
• Electroencephalogram (EEG) signals are crucial for monitoring brain activity and diagnosing epilepsy.
• Accurate and rapid seizure detection is essential for effective patient management and treatment.

Purpose of the Study:

• To develop and validate a model-based classification method for detecting epileptic seizures from EEG signals.
• To utilize the two-point central difference algorithm for enhancing epileptic signal waveforms in EEG data.
• To employ a random forest algorithm for classifying seizure and non-seizure events based on statistical model fitting.

Main Methods:

• A novel EEG filter was designed using the two-point central difference algorithm to enhance epileptic signal characteristics.
• The filtered EEG signals were fitted to a quadratic linear-parabolic model via curve fitting techniques.
• Four statistical parameters derived from model fitting were used as features for a random forest classifier.
• The method was evaluated on 66 epochs from the Children Hospital Boston epilepsy database.

Main Results:

• The proposed method demonstrated fast and accurate detection of epileptic seizures.
• The classification achieved a sensitivity of 92%, a specificity of 96%, and an overall accuracy of 94.1%.
• The model-based approach effectively discriminated between seizure and non-seizure EEG epochs.

Conclusions:

• The developed model-based classification method shows significant promise for real-time epileptic seizure detection.
• The integration of the two-point central difference algorithm and random forest classification offers a robust approach for EEG analysis.
• This technique provides a valuable tool for improving the diagnosis and monitoring of epilepsy.