Related Experiment Video
Updated: Nov 19, 2025
Preterm EEG: A Multimodal Neurophysiological Protocol
Published on: February 18, 2012
Deep Learning for EEG Seizure Detection in Preterm Infants
Alison O'Shea1, Rehan Ahmed1, Gordon Lightbody1
1Irish Centre for Maternal and Child Health Research (INFANT), Department of Electrical and Electronic Engineering, University College Cork, Cork, Ireland.
Insights
Deep learning models improve neonatal seizure detection in preterm infants. Transfer learning from term infant data significantly enhances accuracy, overcoming challenges with limited preterm data.
Area of Science:
- Neonatal neurology
- Medical artificial intelligence
- Signal processing
Background:
- Electroencephalography (EEG) is crucial for neonatal seizure detection, but interpretation in preterm infants is challenging due to differing morphology and scarce expert availability.
- Preterm infants have a higher seizure incidence, necessitating specialized detection algorithms, yet annotated preterm EEG data is limited.
- Existing seizure detection algorithms trained on term infant EEG perform poorly on preterm infants.
Purpose of the Study:
- To explore and compare novel deep learning (DL) architectures for neonatal seizure detection specifically in preterm infants.
- To evaluate different training strategies, including using term infant data, preterm infant data, age-specific preterm data, and transfer learning.
- To address the challenge of limited annotated preterm EEG data for algorithm development.
Main Methods:
- The study utilized a large database of 575 hours of continuous EEG recordings from preterm infants.
- Compared a Support Vector Machine (SVM) classifier trained on term infant EEG with DL approaches.
- Investigated DL models trained on term data, preterm data, and employed transfer learning from a term-trained model.
Main Results:
- A term-trained SVM algorithm showed significantly lower performance (AUC 88.3%) on preterm infants compared to term infants (AUC 96.6%).
- Re-training the SVM on preterm data yielded only a marginal improvement (AUC 89.7%).
- A DL approach achieved a higher initial AUC (93.3%) on preterm data and reached 95.0% using transfer learning, demonstrating effectiveness with minimal preterm data.
Conclusions:
- Deep learning architectures, particularly with transfer learning, offer a promising solution for accurate neonatal seizure detection in preterm infants.
- Transfer learning effectively leverages existing models trained on term infant data, overcoming the scarcity of annotated preterm EEG data.
- The proposed DL approach reduces the need for manual feature engineering and improves detection accuracy in a challenging clinical population.
Abstract:
EEG is the gold standard for seizure detection in the newborn infant, but EEG interpretation in the preterm group is particularly challenging; trained experts are scarce and the task of interpreting EEG in real-time is arduous. Preterm infants are reported to have a higher incidence of seizures compared to term infants. Preterm EEG morphology differs from that of term infants, which implies that seizure detection algorithms trained on term EEG may not be appropriate. The task of developing preterm specific algorithms becomes extra-challenging given the limited amount of annotated preterm EEG data available. This paper explores novel deep learning (DL) architectures for the task of neonatal seizure detection in preterm infants. The study tests and compares several approaches to address the problem: training on data from full-term infants; training on data from preterm infants; training on age-specific preterm data and transfer learning. The system performance is assessed on a large database of continuous EEG recordings of 575[Formula: see text]h in duration. It is shown that the accuracy of a validated term-trained EEG seizure detection algorithm, based on a support vector machine classifier, when tested on preterm infants falls well short of the performance achieved for full-term infants. An AUC of 88.3% was obtained when tested on preterm EEG as compared to 96.6% obtained when tested on term EEG. When re-trained on preterm EEG, the performance marginally increases to 89.7%. An alternative DL approach shows a more stable trend when tested on the preterm cohort, starting with an AUC of 93.3% for the term-trained algorithm and reaching 95.0% by transfer learning from the term model using available preterm data. The proposed DL approach avoids time-consuming explicit feature engineering and leverages the existence of the term seizure detection model, resulting in accurate predictions with a minimum amount of annotated preterm data.
