EEG-Based Machine Learning in Pediatric Epilepsy Research: Worldwide Patterns, Recent Trends and Future Prospects: A Bibliometric Study

Epilepsy is a common chronic neurological disease affecting approximately 65 million people worldwide, with almost 2 million incident cases each year. It is a disorder that has a different clinical appearance in children and a different diagnostic and therapeutic context due to immature neurodevelopment and the clinical and etiological variability of pediatric seizures. The most important is early and accurate diagnosis, as continued seizure activity is known to be associated with irreversible neuronal damage, cognitive deficits, and psychosocial complications. Notably, approximately 30% of pediatric cases are resistant to treatment with standard antiepileptic medications and surgery, highlighting the critical need for the development of novel diagnostic and prognostic tools. Although Electroencephalography (EEG) is still the foundation of the diagnosis and monitoring of epilepsy, traditional manual analysis is highly subjective, time consuming and prone to interrater variability. The advent of Machine Learning (ML) and Deep Learning (DL) architectures has revolutionized EEG analysis and has made it possible to achieve high-throughput, reproducible seizure detection and prediction for pooled sensitivities and specificities of >80% and a DL validation accuracy of 89-91%. This bibliometric review is systematic and uses keyword co-occurrence networks, temporal keyword evolution mapping, density visualizations and thematic clustering to analyze the literature from 2015-2025. Following an increase in numbers after 2014, publication output has now picked up considerably, with 261 publications and 6,124 citations in 2022. The seizure detection accuracy was shown to be 92-97% for the United States and China as the dominant contributors, and hybrid architectures with CNN attention provided the best results. For each research paper, the research field was identified via thematic analysis, which revealed three main clusters: seizure detection, signal processing, and multimodal neuroimaging integration. There remain key gaps, including in pediatric-specific datasets, model interpretation, and clinical translation, that require frameworks and structured interdisciplinary collaboration to ensure the responsible implementation of explainable AI. Further studies should focus on the design of explainable AI frameworks and collaborative cross-institutional validation for the development of personalized, data-driven pediatric epilepsy care.