Machine learning-based lateralization and localization of seizure onset in focal cortical dysplasia patients using ictal scalp EEG.

INRODUCTION: This study aimed to develop and evaluate a machine learning framework for classifying seizure onset zone lateralization and localization in patients with focal cortical dysplasia using ictal scalp electroencephalography (EEG). METHODS: We retrospectively analyzed ictal scalp EEG recordings from 69 patients with focal cortical dysplasia, including 63 surgical and 6 non-surgical patients. EEG signals were preprocessed using common average referencing, segmented into overlapping 1-s windows, and filtered into five frequency bands. Morphological and connectivity features were extracted, and principal component analysis was applied for dimensionality reduction. Automated machine learning was used to select optimal classifiers for lateralization and localization. Model performance was assessed using three-fold cross-validation in 51 surgical patients, internal validation in 12 surgical patients, and extra validation in 6 non-surgical patients. RESULTS: Before principal component analysis, connectivity features generally outperformed morphological features. Covariance-based connectivity achieved the highest area under the receiver operating characteristic curve for lateralization (0.781), whereas the full connectivity feature set achieved the highest area under the receiver operating characteristic curve for localization (0.786). After principal component analysis, morphology-based energy features showed improved performance, achieving area under the receiver operating characteristic curve values of 0.811 for lateralization and 0.829 for localization in the early post-onset window. DISCUSSION: These findings suggest that ictal scalp EEG combined with machine learning enables accurate and interpretable seizure onset zone classification in patients with focal cortical dysplasia. The proposed framework may serve as a noninvasive decision-support tool to improve presurgical evaluation and guide invasive EEG planning in focal cortical dysplasia-related epilepsy.