Principal component analysis-enhanced cosine radial basis function neural network for robust epilepsy and seizure...

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence

Background:

• Accurate classification of electroencephalogram (EEG) signals is crucial for diagnosing neurological disorders like epilepsy.
• Existing methods face challenges in reliably distinguishing between healthy, ictal (seizure), and interictal (between seizure) EEG states.

Purpose of the Study:

• To develop and validate a novel, highly accurate, and robust two-stage classifier for EEG signal analysis.
• To improve the classification of EEGs into healthy, ictal, and interictal states using advanced signal processing and machine learning techniques.

Main Methods:

• A two-stage classifier integrating principal component analysis (PCA) for feature enhancement and a cosine radial basis function neural network (RBFNN) for classification.
• Utilized a previously developed mixed-band wavelet-chaos methodology with a nine-parameter feature space as input.
• Conducted extensive parametric and sensitivity analyses to validate performance and robustness.

Main Results:

• The PCA-enhanced RBFNN classifier achieved an overall EEG classification accuracy of 96.6% with high robustness (1.4% standard deviation).
• When differentiating only normal and interictal EEGs for epilepsy diagnosis, the model reached an accuracy of 99.3%.
• This performance surpasses the detection rates of even highly trained neurologists for interictal EEGs.

Conclusions:

• The proposed wavelet-chaos-neural network methodology offers a significant advancement in automated EEG classification.
• The PCA-enhanced two-stage classifier demonstrates superior accuracy and robustness, particularly valuable for epilepsy diagnosis.
• This advanced computational approach holds promise for improving the early and accurate detection of epilepsy.