Automatic epileptic seizure classification in multichannel EEG time series with linear discriminant analysis

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) is crucial for epilepsy detection.
• Current methods often rely on limited single-channel data.
• A need exists for robust multichannel EEG analysis.

Purpose of the Study:

• To develop a strategy for classifying multichannel EEGs.
• To differentiate between healthy, interictal, and ictal states in epilepsy patients.
• To enhance diagnostic accuracy for neurological disorders.

Main Methods:

• Maximal Overlap Discrete Wavelet Transform (MODWT) for feature extraction.
• Calculation of variance, Pearson correlation, Hoeffding's D, Shannon entropy, and IQR.
• Linear Discriminant Analysis (LDA) for classification.
• Validation on data from 34 healthy individuals and 51 epilepsy patients.

Main Results:

• High accuracy (96.88%) distinguishing healthy from epileptic EEGs (AUC=1).
• Accurate classification (94.12%) between interictal and ictal states (AUC=0.97).
• Excellent discrimination (AUC=1) between interictal EEGs and normal EEGs.
• Achieved 85.88% accuracy and 0.83 AUC for three-class discrimination.

Conclusions:

• The proposed multichannel EEG analysis method demonstrates high efficacy.
• This approach can serve as a valuable auxiliary tool for clinicians.
• It has the potential to reduce the diagnostic burden in epilepsy detection.