Seizure onset zone identification using phase-amplitude coupling and multiple machine learning approaches for...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Neuroscience

Background:

• Epilepsy diagnosis relies on accurate seizure onset zone (SOZ) localization.
• Medically refractory epilepsy necessitates precise SOZ identification for effective treatment.
• Interictal electrocorticogram (ECoG) offers insights into brain activity between seizures.

Purpose of the Study:

• To investigate phase-amplitude coupling (PAC) characteristics in interictal ECoG.
• To evaluate PAC as a potential biomarker for SOZ identification.
• To develop machine learning models for automated SOZ classification using PAC features.

Main Methods:

• Calculated PAC between low-frequency rhythms (0.5-24 Hz) and high-frequency oscillations (HFOs) (80-560 Hz) using the mean vector length modulation index on 20-s ECoG windows.
• Employed statistical measures to compare PAC between SOZ and non-seizure onset zones (NSOZ).
• Developed novel machine learning models with feature-wise/class-wise re-weighting strategies and time-series nested cross-validation for classification and evaluation.

Main Results:

• Demonstrated significant PAC differences between SOZ and NSOZ, particularly at specific slow wave and HFO band pairs.
• Validated the effectiveness of PAC features for SOZ identification.
• Achieved improved classification performance using the proposed machine learning models.

Conclusions:

• PAC is a promising biomarker for SOZ identification in interictal ECoG.
• Machine learning models effectively leverage PAC features for automated SOZ classification.
• The findings contribute to improved diagnosis and treatment strategies for epilepsy.