Related Experiment Video
Updated: May 5, 2026
08:20
Brain Source Imaging in Preclinical Rat Models of Focal Epilepsy using High-Resolution EEG Recordings
Published on: June 6, 2015
15.3K
Deep Learning for Generalized EEG Seizure Detection after Hypoxia-Ischemia-Preclinical Validation
Hamid Abbasi1,2, Joanne O Davidson1, Simerdeep K Dhillon1
1Department of Physiology, Faculty of Medical and Health Sciences, University of Auckland, Auckland 1023, New Zealand.
Bioengineering (Basel, Switzerland)
|March 27, 2024
Summary
Advanced deep learning accurately detects neonatal seizures after hypoxia-ischemia (HI), even with brain cooling. This convolutional neural network (CNN) approach shows high reliability across different brain maturities for clinical translation.
Area of Science:
- Neonatal neurology
- Computational neuroscience
- Medical signal processing
Background:
- Neonatal seizures following hypoxia-ischemia (HI) are influenced by brain maturity and therapeutic hypothermia (TH).
- Accurate, generalized automatic seizure identification is crucial for clinical management.
Purpose of the Study:
- To validate deep learning (CNN) classifiers for detecting seizures post-HI in fetal sheep.
- To assess the impact of brain maturation and cooling on seizure detection accuracy.
Main Methods:
- Utilized convolutional neural networks (CNNs) for seizure detection in EEG data from fetal sheep.
- Employed leave-one-out and k-fold cross-validation on cohorts including HI-normothermia term, HI-hypothermia term, and HI-normothermia preterm groups (>17,300 hours of recordings).
Main Results:
- Term-trained detectors achieved high accuracy (99.5%) for preterm HI seizures.
- Preterm training data led to decreased performance on term and hypothermia data.
- Overall average accuracy of 99.7% (AUC 99.4%) was achieved across all detectors.
Conclusions:
- Deep learning algorithms reliably identify post-HI seizures irrespective of maturity, with minimal impact from hypothermia.
- Seizure spectral features differ between preterm and term neonates.
- This advancement offers a clinically translatable tool for seizure detection in 256Hz EEG recordings.
