Using Explainable AI for EEG-based Reduced Montage Neonatal Seizure Detection

Addresses a critical global health equity gap: The paper tackles neonatal seizure detection in resource-limited settings by reducing the number of EEG electrodes needed, potentially making life-saving diagnosis accessible beyond wealthy hospitals with advanced technology and specialized staff.

Transparency built into the AI system: Unlike many "black box" AI models, this system provides real-time interpretability, meaning clinicians can understand why the AI flagged a potential seizure rather than simply trusting an unexplained diagnosis—crucial for medical decision-making.

Significant performance improvements: The model achieved a 42.86% improvement in recall (catching actual seizures), which is particularly important in neonatal care where missing a seizure could have severe developmental consequences for vulnerable infants.

Automation could reduce diagnostic delays: By automating detection and reducing dependence on continuous expert monitoring, this approach could enable faster diagnosis and treatment initiation, especially important since early intervention in neonatal seizures is critical for brain health outcomes.

Practical design for real-world deployment: The use of a reduced EEG montage (fewer electrodes) and real-time processing capability suggests this technology could realistically be implemented in under-resourced NICUs, not just in research settings.