Seizure prediction using polynomial SVM classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy affects millions globally, necessitating accurate seizure prediction for improved patient care and safety.
• Current seizure prediction methods often require significant computational resources, limiting their clinical applicability.
• Developing low-complexity, low-power algorithms is crucial for real-time, patient-specific seizure forecasting.

Purpose of the Study:

• To develop and validate a novel patient-specific algorithm for predicting seizures in epileptic patients.
• To achieve high prediction accuracy with reduced hardware complexity and low power consumption.
• To compare the performance of the proposed algorithm against a baseline approach using a standard epilepsy database.

Main Methods:

• Computed spectrograms from fragmented electroencephalogram (EEG) signals.
• Extracted band powers, relative spectral powers, and ratios of spectral powers as features.
• Employed electrode and feature selection using classification and regression trees.
• Utilized radial basis function kernel support vector machine (RBF-SVM) for baseline and polynomial SVM (degree 2) for the proposed method.
• Tested the algorithm on intra-cranial EEG (iEEG) data from the American Epilepsy Society Seizure Prediction Challenge database.

Main Results:

• The baseline experiment achieved 100% sensitivity and an average Area Under Curve (AUC) of 0.9985.
• The proposed algorithm, using fewer features and a polynomial SVM, achieved 100% sensitivity and an average AUC of 0.9795.
• Both experiments utilized only 10% of the available training data, demonstrating efficiency.

Conclusions:

• The proposed patient-specific algorithm offers a viable solution for real-time seizure prediction with reduced computational load.
• The method demonstrates high sensitivity and good AUC, making it a promising tool for epilepsy management.
• This approach has the potential for integration into wearable devices for continuous epilepsy monitoring.