Noninvasive Preoperative Evaluation in Drug-Resistant Epilepsy Based on the Brain Network Topological Dynamics

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Drug-resistant epilepsy (DRE) impacts millions globally, with surgery offering the best treatment but only succeeding in about 50% of patients.
• Accurate prediction of surgical outcomes is crucial to prevent unnecessary invasive intracranial monitoring and optimize patient selection.
• Current prediction methods often rely on invasive procedures, highlighting the need for noninvasive alternatives.

Purpose of the Study:

• To develop and validate a noninvasive framework for predicting surgical outcomes in DRE patients.
• To identify reliable brain network topological features from scalp EEG as prognostic markers.
• To leverage machine learning for personalized preoperative evaluation.

Main Methods:

• A retrospective study involving 43 DRE patients and 110 seizures.
• Integration of scalp EEG source imaging, brain network topological features (GEDIFF, KDEN, T, BC, STR), and machine learning algorithms (KNN, CatBoost, ExtraTrees).
• Leave-one-out and five-fold cross-validation were employed to assess model performance.

Main Results:

• Global Diffusion Efficiency (GEDIFF) and Network Density (KDEN) were identified as robust prognostic markers (p < 0.001).
• Machine learning models achieved high accuracy (up to 86.4%), precision (up to 90.1%), and AUC (94.6%).
• Noninvasive model performance was comparable to invasive iEEG-based methods using standard 16-channel scalp EEG.

Conclusions:

• Favorable surgical outcomes in DRE are linked to more isolated epileptogenic zone networks.
• Noninvasive brain network analysis using scalp EEG shows significant clinical potential for personalized preoperative surgical evaluation.
• The developed framework offers a reliable, noninvasive alternative for predicting DRE surgery success.