Tensor-based Uncorrelated Multilinear Discriminant Analysis for Epileptic Seizure Prediction

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Epileptic seizure prediction offers a critical time advantage over traditional detection methods, enabling timely patient treatment.
• Accurate forecasting of seizure onset remains a significant challenge in epilepsy management.

Purpose of the Study:

• To develop and evaluate a novel spectral feature extraction method for enhanced epileptic seizure prediction.
• To improve the accuracy and reliability of seizure forecasting using advanced tensor-based analysis and machine learning algorithms.

Main Methods:

• A three-order tensor in temporal, spectral, and spatial domains was constructed using wavelet transform for comprehensive data representation.
• Uncorrelated Multilinear Discriminant Analysis (UMLDA) was employed for tensor-to-vector projection (TVP) with minimum redundancy.
• Support Vector Machine (SVM) was utilized for the classification of ictal and preictal states, using EEG data from 23 subjects.

Main Results:

• The proposed method demonstrated a high overall accuracy of 95% in classifying ictal and preictal states.
• The tensor-based approach effectively leveraged multi-dimensional information from EEG signals for improved prediction performance.

Conclusions:

• The developed spectral feature extraction technique combined with UMLDA and SVM shows significant promise for accurate epileptic seizure prediction.
• This approach offers a valuable tool for improving patient care by providing advanced warning of epileptic seizures.