Bispectrum Features and Multilayer Perceptron Classifier to Enhance Seizure Prediction

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Drug-refractory epilepsy significantly impacts patient quality of life.
• Accurate seizure forecasting requires identifying precursors from electroencephalography (EEG) recordings.
• Existing EEG features struggle to independently characterize pre-seizure brain states.

Purpose of the Study:

• To assess the feasibility of using bispectrum analysis, a higher-order statistics technique, as a seizure activity precursor.
• To identify quantitative features from the bispectrum that differentiate preictal and interictal states.
• To evaluate the predictive performance of bispectrum-derived features using a machine learning classifier.

Main Methods:

• Extracted quantitative features from the bispectrum of EEG recordings.
• Applied statistical tests to identify significant differences between preictal and interictal states.
• Utilized normalized bispectral entropy, normalized bispectral squared entropy, and mean magnitude as inputs for a multilayer perceptron classifier.

Main Results:

• All bispectrum-extracted features showed statistically significant differences (p < 0.05) between preictal and interictal states.
• The multilayer perceptron classifier achieved accuracies of 78.11% (normalized bispectral entropy), 72.64% (normalized bispectral squared entropy), and 73.26% (mean magnitude).

Conclusions:

• Bispectrum analysis is a feasible and effective method for identifying seizure precursors in EEG.
• Bispectrum-derived features hold significant potential for improving the accuracy of seizure prediction algorithms.
• This approach offers a promising avenue for enhancing the quality of life for individuals with epilepsy.