Related Experiment Video
Updated: Dec 7, 2025
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
Attention-Based Network for Weak Labels in Neonatal Seizure Detection
Dmitry Yu Isaev1, Dmitry Tchapyjnikov2, C Michael Cotten3
1Department of Biomedical Engineering, Duke University, Durham, NC, USA.
Insights
Detecting neonatal seizures with deep learning is challenging due to imbalanced data and localized activity. This study evaluates models and proposes a channel importance mechanism for improved seizure detection in neonatal intensive care units (NICUs).
Area of Science:
- Clinical Neurophysiology
- Artificial Intelligence in Medicine
- Neonatal Medicine
Background:
- Seizures are a frequent complication in newborns undergoing therapeutic hypothermia for hypoxic ischemic encephalopathy, necessitating continuous electroencephalographic (EEG) monitoring.
- Current intermittent review of EEG data leads to delays in seizure detection and treatment, highlighting the need for automated solutions.
- Deep learning approaches for neonatal seizure detection face challenges including data imbalance and the time-consuming nature of channel-specific annotations.
Purpose of the Study:
- To assess the impact of different deep learning models and data balancing techniques on neonatal seizure detection from EEG.
- To propose and evaluate a novel deep learning model that assigns importance levels to EEG channels, acting as a proxy for seizure activity.
- To provide a preliminary comparison of deep learning model performance against human expert decisions for clinical relevance.
Main Methods:
- Evaluation of various deep learning architectures and data balancing strategies for neonatal seizure detection using EEG.
- Development and testing of a model incorporating a channel importance mechanism to better localize and identify seizure activity.
- Quantitative assessment of the proposed model's performance and its portability across different EEG device layouts without retraining.
Main Results:
- The study quantitatively assessed the effectiveness of the channel importance mechanism in deep learning-based neonatal seizure detection.
- The proposed model demonstrated portability across different EEG device configurations, reducing the need for retraining.
- High Area Under the Curve (AUC) values in deep learning models did not consistently correlate with agreement with human expert raters.
Conclusions:
- Deep learning models show promise for neonatal seizure detection, but challenges related to data imbalance and channel-specific activity remain.
- The developed channel importance mechanism offers a potential improvement for automated seizure detection and clinical deployment.
- Further refinement of deep learning algorithms is crucial to ensure optimal seizure discrimination and alignment with expert clinical judgment.
Abstract:
Seizures are a common emergency in the neonatal intesive care unit (NICU) among newborns receiving therapeutic hypothermia for hypoxic ischemic encephalopathy. The high incidence of seizures in this patient population necessitates continuous electroencephalographic (EEG) monitoring to detect and treat them. Due to EEG recordings being reviewed intermittently throughout the day, inevitable delays to seizure identification and treatment arise. In recent years, work on neonatal seizure detection using deep learning algorithms has started gaining momentum. These algorithms face numerous challenges: first, the training data for such algorithms comes from individual patients, each with varying levels of label imbalance since the seizure burden in NICU patients differs by several orders of magnitude. Second, seizures in neonates are usually localized in a subset of EEG channels, and performing annotations per channel is very time-consuming. Hence models which make use of labels only per time periods, and not per channels, are preferable. In this work we assess how different deep learning models and data balancing methods influence learning in neonatal seizure detection in EEGs. We propose a model which provides a level of importance to each of the EEG channels - a proxy to whether a channel exhibits seizure activity or not, and we provide a quantitative assessment of how well this mechanism works. The model is portable to EEG devices with differing layouts without retraining, facilitating its potential deployment across different medical centers. We also provide a first assessment of how a deep learning model for neonatal seizure detection agrees with human rater decisions - an important milestone for deployment to clinical practice. We show that high AUC values in a deep learning model do not necessarily correspond to agreement with a human expert, and there is still a need to further refine such algorithms for optimal seizure discrimination.
