Authors :
Presenting Author: Youmin Shin, MS – Seoul National University Hospital
Sungeun Hwang, MD – Ewha Womans University Mokdong Hospital
Hyoshin Son, MD – St. Mary's Hospital
SangKun Lee, MD, PhD – Seoul National University
Young-Gon Kim, PhD – Seoul National University Hospital
Kyung-Il Park, MD, PhD – Seoul National University
Rationale:
Focal cortical dysplasia (FCD) is a leading cause of drug-resistant epilepsy, where surgical outcomes critically depend on precise identification of the seizure onset zone (SOZ). Ictal scalp EEG offers a noninvasive alternative that captures the real-time dynamics of seizure initiation. In this study, we propose a machine learning framework that utilizes ictal EEG to classify SOZ lateralization and localization, aiming to support presurgical decision-making in patients with FCD.
Methods:
A retrospective analysis was performed on 69 patients with FCD. Of these, 51 patients (15 left, 36 right; 34 temporal, 17 extratemporal) were used for model development, 12 surgical patients (4 left, 8 right; 7 temporal, 5 extratemporal) for internal validation, and 6 non-surgical patients (2 left, 4 right; 3 temporal, 3 extratemporal) for extra validation. Ictal EEG recordings were preprocessed using baseline correction (−10 to −5 s), common average referencing, and bandpass filtering spanning delta to gamma bands. Seizure onset was annotated by experts, and overlapping 1-second epochs were extracted from −5 to +10 seconds. Morphological (Hjorth parameters, energy, statistical moments) and connectivity (correlation, covariance) features were computed. PCA was applied to isolate variance components most responsive to ictal transitions. Separate binary classifiers were trained using both raw and PCA-transformed features to evaluate their relative effectiveness in SOZ classification.
Results:
Without PCA, connectivity-based features outperformed morphological ones, achieving AUCs of 0.830 for lateralization (internal: 0.813; extra: 0.750) and 0.837 for localization (internal: 0.781; extra: 0.750). With PCA, peak AUCs slightly increased to 0.848 for lateralization (internal: 0.800; extra: 0.833) and 0.863 for localization (internal: 0.786; extra: 0.778) (Figure 1). Before PCA, the most influential features for lateralization involved interhemispheric covariance between electrodes such as Fp2–T7 and T7–Cz. For localization, fronto-parietal and temporo-central connections were most discriminative. After PCA, the most informative components were mid-level axes associated with delta- and beta-band energy entropy, rather than early components dominated by global amplitude or noise (Figure 2).
Conclusions:
This study demonstrates the feasibility and clinical utility of ictal EEG-based machine learning models for localizing and lateralizing seizure onset in FCD. The analysis pipeline, applied to expert-marked ictal scalp EEG segments, integrated frequency-based features and PCA-driven dimensionality reduction to enable robust and interpretable classification. These findings suggest that the proposed framework could augment presurgical planning in drug-resistant epilepsy and support more personalized, data-driven clinical decision-making.
Funding:
This research was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: RS-2023-00265638).