Clinical implementation of a neonatal seizure detection algorithm

Area of Science:

• Biomedical Engineering
• Clinical Neurophysiology

Background:

• Automated neonatal seizure detection systems are advancing towards clinical bedside implementation.
• Effective visualization of seizure detection output is crucial for accurate interpretation and clinical utility.

Purpose of the Study:

• To evaluate different visualization methods for neonatal seizure detection system output.
• To analyze the impact of these visualizations on system performance in a clinical setting.
• To gather clinical personnel feedback on the usefulness and accuracy of presented methods.

Main Methods:

• Comparison of three visualization techniques: binary output, probabilistic trace, and spatio-temporal colormap.
• Inclusion of audified neonatal electroencephalogram (EEG) as an alternative sensory output.
• Conducting a survey among clinical staff to assess visualization method utility and accuracy.

Main Results:

• The developed neonatal seizure detector, with appropriate visualization, can serve as a valuable clinical decision support tool.
• Survey results indicate a preference for a combined binary output and probabilistic trace visualization.
• Established a link between information visualization techniques and the computation of conventional performance metrics.

Conclusions:

• A combined binary and probabilistic trace visualization is recommended for clinical implementation of neonatal seizure detection systems.
• The decision support system shows promise for regulatory approval and clinical adoption.
• The tool offers significant potential healthcare benefits for neonatal seizure management.