Quadratic time-frequency distribution selection for seizure detection in the newborn
N Stevenson1, M Mesbah, B Boashash
1Centre for Clinical Research, University of Queensland, Royal Brisbane, Women's Hospital, Herston 4029, QLD, Australia. n.stevenson@uq.edu.au
Summary
This study identifies the optimal quadratic time-frequency distribution (QTFD) for detecting seizures in newborn EEG. The modified B distribution best distinguishes seizure from non-seizure EEG patterns.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Neonatal Neurology
Background:
- Newborn EEG seizure detection often relies on quadratic time-frequency distributions (QTFDs) for time-frequency representations (TFRs).
- The optimal QTFD for discriminating between seizure and non-seizure EEG epochs remains underexplored.
Purpose of the Study:
- To identify the optimal QTFD for maximizing the difference between seizure and non-seizure EEG TFRs in newborns.
- To provide a data-driven method for selecting QTFDs for neonatal EEG analysis.
Main Methods:
- A data-driven optimization process was employed to select the best QTFD.
- The selection criterion was the maximization of the absolute error between seizure and non-seizure QTFDs.
- Several non-adaptive QTFDs with restricted kernel volumes were compared.
Main Results:
- The modified B distribution, a lag-independent or narrowband QTFD, demonstrated superior performance.
- This QTFD effectively highlighted differences in time-frequency energy between seizure and non-seizure newborn EEG.
- The optimization procedure successfully identified a QTFD suitable for EEG seizure detection.
Conclusions:
- The modified B distribution is recommended as the optimal QTFD for newborn EEG seizure detection.
- This finding can improve the accuracy and reliability of automated seizure detection systems in neonates.
- Further research can explore adaptive QTFDs for enhanced neonatal EEG analysis.
