Artificial Intelligence and Machine Learning in Pediatric Epilepsy: A Systematic Review

To evaluate the progress of artificial intelligence (AI)-based tools in interpreting clinical data, to compare the existing models, to identify the most proficient models and to determine the limitations of current models and existing research.

Three databases were searched to identify studies in any language that met the eligibility criteria. At least three independent reviewers screened and evaluated each record.

Convolutional neural networks emerged as the predominant deep learning architecture, whereas support vector machines were the most frequently used classical machine learning approach. AI-driven seizure detection showed diagnostic performance approaching that of experienced specialists. Reported applications include the detection of specific epilepsy syndromes, identification of electrical status epilepticus during sleep (ESES) and spike-wave index, continuous electroencephalography (EEG) surveillance, neonatal monitoring, and wearable seizure detection technologies.

AI models achieved diagnostic accuracy exceeding 90% in distinguishing normal EEG recordings from abnormal ones, with automated seizure detection representing the most widespread case of clinical use. Despite encouraging results, the reliability of these systems is constrained by limited cohort sizes (typically < 100 patients). Future efforts should focus on large-scale, multicenter validation to enable clinical implementation.

INPLASY database, https://doi.org/10.37766/inplasy2025.10.0020 .