Multiband entropy-based feature-extraction method for automatic identification of epileptic focus based on...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate localization of the epileptic focus is critical for successful epilepsy surgery.
• High-frequency oscillations (HFOs) in high-frequency subbands (>80 Hz) of intracranial electroencephalogram (iEEG) are known biomarkers for seizure onset zones (SOZ).

Purpose of the Study:

• To develop and validate a machine learning approach for detecting the epileptic focus using information theoretic features from high-frequency subbands of interictal iEEG.
• To address the challenge of imbalanced data in identifying SOZ and non-SOZ channels.

Main Methods:

• Decomposition of multi-channel interictal iEEG signals into high-frequency subbands.
• Calculation of entropy features for each subband and selection using sparse linear discriminant analysis (sLDA).
• Application of an adaptive synthetic oversampling approach (ADASYN) with support vector machine (SVM) to handle imbalanced learning and detect focal segments.

Main Results:

• The proposed method successfully identified focal segments indicative of the epileptic focus.
• Experimental results on eight patients demonstrated accurate and efficient detection of the epileptic focus.
• Statistical tests confirmed the efficacy of the automatic detector.

Conclusions:

• The developed machine learning approach effectively utilizes high-frequency subband entropy features for epileptic focus detection.
• The method provides an accurate and efficient tool for presurgical investigation in epilepsy.
• The study highlights the potential of advanced signal processing and machine learning in improving epilepsy diagnosis and treatment planning.