Epileptic Seizures Prediction Using Machine Learning Methods.

Area of Science:

• Neurology
• Computational Neuroscience
• Biomedical Engineering

Background:

• Epileptic seizures result from brain dysfunction, impacting patient health.
• Early prediction of seizures aids in timely medication and prevention.
• Current machine learning approaches for seizure prediction from Electroencephalograms (EEG) face challenges in signal preprocessing and feature extraction, affecting prediction accuracy and anticipation time.

Purpose of the Study:

• To propose a novel model for reliable preprocessing and feature extraction of EEG signals for epileptic seizure prediction.
• To improve the true positive prediction rate and anticipation time for the preictal state.

Main Methods:

• Applied Empirical Mode Decomposition (EMD) for EEG signal preprocessing and noise removal.
• Extracted time and frequency domain features from preprocessed EEG signals.
• Trained a prediction model using the extracted features to detect the preictal state.

Main Results:

• The proposed model achieved a high true positive rate of 92.23% in detecting the preictal state.
• The model demonstrated a maximum anticipation time of 33 minutes and an average prediction time of 23.6 minutes.
• Performance was evaluated on the scalp EEG CHB-MIT dataset comprising 22 subjects.

Conclusions:

• The developed model offers a reliable approach to EEG signal preprocessing and feature extraction for epileptic seizure prediction.
• The model significantly enhances both the accuracy and anticipation time for predicting the onset of epileptic seizures.
• This advancement holds promise for improving patient management and quality of life through more effective seizure forecasting.