Neonatal seizure detection: a review of current methods, challenges and future directions

Abstract Seizures occur more frequently in the neonatal period than at any other time, serving as markers of underlying brain injury and potentially contributing to further neurological damage. Seizures in neonates are predominantly electrographic only, and even when clinical signs appear, these are typically subtle and easily overlooked by clinical observation alone. Additionally, post-treatment uncoupling phenomena further complicate clinical seizure recognition. Amplitude-integrated electroencephalography (aEEG) provides a practical bedside tool for seizure detection, but its utility is limited by susceptibility to artefacts and reduced sensitivity for short-duration or focal seizures. Although conventional EEG monitoring, incorporating continuous video-EEG, remains the gold standard, its widespread implementation is hindered by resource-intensive requirements, specialised equipment, and the need for trained personnel. Recent advances in automated seizure detection, particularly through machine learning and deep learning techniques, have significantly improved accuracy and clinical utility, enabling more timely interventions. Emerging research also supports multimodal monitoring approaches, combining EEG data with additional physiological metrics to enhance seizure prediction and detection. Future research should focus on refining deep learning methodologies and increasing dataset diversity to improve algorithm accuracy and generalisability. Personalised medicine and predictive modelling also hold promise for early identification of at-risk neonates, potentially allowing targeted interventions to improve outcomes. Further advances in multimodal monitoring and artificial intelligence are set to significantly transform neonatal neurocritical care.