FetalSleepNet: A Transfer Learning Framework with Spectral Equalisation Domain Adaptation for Fetal Sleep State
IEEE Journal of Biomedical and Health Informatics
|April 8, 2026
Summary
FetalSleepNet automates fetal sleep staging using deep learning and transfer learning, achieving 86.6% accuracy. This novel approach overcomes data scarcity for improved fetal neurodevelopmental monitoring.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Machine Learning
Background:
- Fetal sleep state classification is crucial for detecting neurodevelopmental issues like hypoxia.
- Manual annotation of fetal electroencephalography (fEEG) is subjective, time-consuming, and limited by scarce data.
Purpose of the Study:
- To develop an automated sleep staging system for fEEG data.
- To address the challenge of limited labeled fEEG data using transfer learning.
Main Methods:
- Proposed FetalSleepNet, a deep learning architecture for automated fEEG sleep staging.
- Implemented a cross-developmental and cross-species transfer learning framework.
- Utilized Spectral Equalisation (SE) to align adult human EEG with fetal sheep EEG frequency characteristics.
Main Results:
- Direct transfer learning achieved only 18.7% accuracy.
- Applying Spectral Equalisation (SE) improved accuracy to 73.6% with a frozen model.
- FetalSleepNet with fine-tuning reached state-of-the-art 86.6% accuracy and 62.5% macro F1-score.
Conclusions:
- FetalSleepNet demonstrates effective automated fetal sleep staging.
- The 'label engine' paradigm facilitates training on non-invasive modalities.
- Enables scalable, real-time fetal monitoring and early risk prediction.
