12 - Testing the NeuroBell Luna Neonatal Seizure Detection Algorithm in a Real-World Setting

Publication Number: 1010.12

Aisling Fanning, Infant Research Centre, University College Cork, Cork, Cork, Ireland; Sergio Morant Galvez, Universitat de Valencia, Picanya, Comunidad Valenciana, Spain; Andreea M. Pavel, Cork University Maternity Hospital, Cork, Cork, Ireland; Anne Murray, University College Cork, Cork, Cork, Ireland; Mark O'Sullivan, NeuroBell, Cork, Cork, Ireland; Brian H. Walsh, INFANT Research Centre, Cork, Cork, Ireland; Alison O'Shea, NeuroBell, Ballincollig, Cork, Ireland

Background: The NeuroBell Luna Neonatal Seizure Detection Algorithm (NSDA) is intended to mark sections of EEG that correspond to electrographic seizures in term neonates. In this study the algorithm was tested on a prospective cohort in Cork University Maternity Hospital.

Objective: To quantify the performance of the NSDA in comparison with the electrographic seizures identified by a clinical expert annotator.

Design/Methods: All neonates (36-44 weeks gestation) requiring a multi-channel EEG recording for any reason were eligible to participate with parental consent. All infants enrolled in the study were monitored using conventional EEG (Deymed TruScan), only the EEG recordings from the conventional EEG monitor were used to inform clinical care. The NeuroBell Luna NSDA was applied to 8-channel EEG recorded with the NeuroBell EEG Amplifier in parallel with the conventional EEG monitor, at 250Hz sampling rate.
EEG recordings from all infants were processed through the NeuroBell Luna NSDA post-hoc. Any infant diagnosed with seizures had their EEG reviewed by an expert reviewer and all electrographic seizures captured during the NeuroBell recordings were annotated. An electrographic seizure was defined as a ≥10-second evolving and rhythmic pattern.
The NSDA performance is described by the rate of falsely detected seizures (False Detections per Hour:FD/h) and the number of correctly detected seizures (Positive Percent Agreement: PPA). A correctly detected seizure is determined if there is any overlap between the annotation of the NSDA and the expert reviewer.
The Office of National Research Ethics Committees of Ireland approved the study.

Results: Seventeen infants were enrolled from November 2024 to March 2025. Infant demographics are summarised in Table 1. Approximately 377 hours of EEG were captured by the NeuroBell EEG Amplifier. Four infants had electrographic seizures during their admission. However, only one infant had seizures captured while connected to the NeuroBell EEG device. All 4 electrographic seizures annotated were correctly detected by the NeuroBell Luna NSDA (Figure 1). The longest seizure was 169 seconds in duration and the shortest was 45 seconds. Across all 377 EEG hours, 5 false detections were generated by the NSDA, resulting in 1 false detection every 77 hours (FD/h of 0.013).

Conclusion(s): We have demonstrated a good performance of the NeuroBell Luna NSDA on a small, unseen, prospective cohort of neonates. The low number of seizures captured is a limitation of the study but the low number of false detections in many hours of EEG recordings, demonstrates the specificity of the algorithm.

Table 1: Infant Demographics
IQR: Interquartile Range, CPAP: Continuous Positive Airway Pressure, HIE: Hypoxic-Ischaemic Encephalopathy
*Data not available for 3 infants
**Data not available for 6 infants

Figure 1
Example of electrographic seizure detected by NSDA (purple) and correlation with expert's annotations (blue).