Improvement of an automated neonatal seizure detector using a post-processing technique

Area of Science:

• Biomedical Engineering
• Neuroscience
• Neonatal Medicine

Background:

• Visual seizure detection in neonatal intensive care units (NICUs) is resource-intensive and requires specialized expertise.
• Automated seizure detection systems are needed to aid clinical decisions and reduce workload, but face challenges due to non-stationary neonatal EEG signals and artifacts.
• Previous heuristic rule-based detectors require improvement for clinical utility.

Purpose of the Study:

• To enhance an existing neonatal seizure detection system.
• To improve detection accuracy and reduce the false alarm rate in neonatal electroencephalogram (EEG) monitoring.
• To support clinical decision-making and alleviate workload in NICUs.

Main Methods:

• Incorporation of mean phase coherence as a novel feature to characterize EEG artifacts.
• Application of a support vector machine for post-processing to eliminate false positive detections.
• Evaluation of an improved neonatal seizure detection algorithm based on heuristic if-then rules.

Main Results:

• The false alarm rate was reduced by 42%, decreasing from 2.6 to 1.5 per hour.
• The good seizure detection rate experienced a minimal reduction of only 4%.
• The enhanced system demonstrates improved performance in distinguishing neonatal seizures from artifacts.

Conclusions:

• The integration of mean phase coherence and support vector machine significantly enhances neonatal seizure detection accuracy.
• The improved system effectively reduces false alarms, making it a more reliable tool for clinical application.
• This advancement offers a promising solution for continuous EEG monitoring in NICUs, aiding clinicians and improving patient care.