| Title: |
Classification of Hypoxic Ischemic Encephalopathy (HIE) in Neonates Using ECG and a Transformer-Based Model |
| Authors: |
Kimia Rezaei; Sean R. Mathieson; Gordon Lightbody; Geraldine B. Boylan; William P. Marnane |
| Source: |
IEEE Access, Vol 14, Pp 24298-24311 (2026) |
| Publisher Information: |
IEEE, 2026. |
| Publication Year: |
2026 |
| Collection: |
LCC:Electrical engineering. Electronics. Nuclear engineering |
| Subject Terms: |
ECG; HIE; multiscale tokenization; spectrogram; transformer; neonatal; Electrical engineering. Electronics. Nuclear engineering; TK1-9971 |
| Description: |
Hypoxic-Ischemic Encephalopathy (HIE) is caused by inadequate blood flow and oxygen delivery to the brain, typically occurring around the time of birth. In the Neonatal Unit, neonates with moderate to severe HIE can be treated by employing therapeutic hypothermia. The increasing use of electroencephalogram (EEG) for supporting clinical decisions, such as initiating therapeutic hypothermia, seizure detection and prognostication, led to growing interest in developing automated systems to assist clinicians in grading HIE-EEG. Electrocardiogram (ECG) signals offer a potential alternative to EEG, being widely accessible and routinely recorded. This study investigates three different approaches for binary classification of infant HIE, utilizing a large ECG dataset from neonates. The traditional method involves deriving the heart rate (HR) from the ECG signal, followed by feature extraction in both time and frequency domains, and classification using classical machine learning techniques such as XGBoost. In contrast, two new approaches are proposed: one extracts spectrograms directly from the ECG signal to input into a convolutional neural network (CNN), while the other uses the raw ECG signal as input to transformer-based models. This paper proposes Trend-Augmented Multiscale Tokenization Transformer (TMFormer), a transformer-based model that introduces a tokenization approach derived from raw ECG signal attributes, capturing spatial patterns at multiple scales and tracking the signal’s long-term progression. Additionally, the transformer’s output on the ECG is analyzed in relation to the corresponding NN intervals. The proposed TMFormer using raw ECG outperform a standard transformer and other models using HRV based features and ECG spectrogram. By eliminating the ECG preprocessing step and leveraging more robust deep learning models, computational overhead can be reduced, enabling real-time processing, and enhancing the practicality of supporting clinicians in desicion-making regarding the HIE binary category even in resource-limited hospital settings. |
| Document Type: |
article |
| File Description: |
electronic resource |
| Language: |
English |
| ISSN: |
2169-3536 |
| Relation: |
https://ieeexplore.ieee.org/document/11381441/; https://doaj.org/toc/2169-3536 |
| DOI: |
10.1109/ACCESS.2026.3662271 |
| Access URL: |
https://doaj.org/article/242b2eab2d2d4bf98f9e4d0b487dfea1 |
| Accession Number: |
edsdoj.242b2eab2d2d4bf98f9e4d0b487dfea1 |
| Database: |
Directory of Open Access Journals |