HRVConformer: Deep Learning for HIE Classification from Heart Rate Signals
A new deep learning architecture called HRVConformer classifies hypoxic-ischemic encephalopathy (HIE) using raw instantaneous heart rate signals. Unlike traditional methods that rely on handcrafted features, HRVConformer processes signals end-to-end with a hybrid Convolution-Transformer framework. Convolutional layers extract local features, while Transformer attention mechanisms capture global context. The model was trained on a large dataset of 1,573 one-hour epochs, including 259 expert-annotated epochs and weakly labeled data. A 314-hour validation set provided robust performance estimation, with an independent 215-hour test set used for evaluation. The study is published on arXiv (2605.26190).
Key facts
- HRVConformer is a novel deep learning architecture for HIE classification.
- It uses raw instantaneous heart rate signals in an end-to-end manner.
- The model combines convolutional layers and Transformer attention mechanisms.
- Trained on 1,573 one-hour epochs, including 259 expert-annotated epochs.
- Validation set of 314 hours; independent test set of 215 hours.
- Published on arXiv with ID 2605.26190.
- Avoids handcrafted features by directly processing raw HR signals.
- Captures both local and long-range dependencies in the signal.
Entities
Institutions
- arXiv