Artificial intelligence–assisted video-eeg analysis for localization of epileptogenic foci in MRI-positive pediatric epilepsy and its correlation with neuroimaging findings

Objective: The primary objective of this study is to non-invasively and automatically predict the epileptogenic zone by analyzing time- and frequency-domain features extracted from 24-hour video-EEG monitoring data of pediatric epilepsy patients with structural lesions confirmed on brain MRI. It is hypothesized that AI-assisted EEG analysis may reduce inter-observer variability in visual interpretation and contribute to the localization of epileptic foci. Materials and Methods: A total of 32 children were included in the study, consisting of 22 pediatric epilepsy patients with brain MRI-confirmed structural lesions and 10 healthy controls with normal EEG/MRI findings. Time- and frequency-domain features such as power spectral density calculated using the Welch method, absolute band powers, Hjorth mobility and complexity parameters, and mean time-domain skewness were extracted from the EEG signals. These features were then used to train Decision Tree, Random Forest, and XGBoost classifiers, and the models were compared in terms of performance metrics. Results: Among the tested models, XGBoost achieved the highest diagnostic performance with an accuracy of 75%, sensitivity of 80%, and specificity of 66%. Hjorth parameters and time-domain asymmetry measures demonstrated significant discriminative value, consistent with the lesion side. Source distribution analyses also revealed increased topographic activity in the hemisphere ipsilateral to the lesion, showing concordance with brain MRI findings. Notably, in some patients with no evident epileptic activity on V-EEG, AI-based analysis detected electrophysiological abnormalities in the lesional region. Conclusion: This study demonstrates that AI-assisted analysis of long-term EEG recordings can non-invasively identify the epileptogenic zone in pediatric epilepsy. In addition, detecting subclinical activity and secondary foci discordant with lesion localization may support clinical decision-making. Larger-scale studies incorporating multimodal data and deep learning approaches are expected to further enhance the clinical applicability of this method. Keywords: Pediatric epilepsy, Video-EEG, Artificial intelligence, Machine learning, Hjorth parameters.