Related Experiment Video
Updated: Sep 28, 2025
07:40
Multi-Modal Home Sleep Monitoring in Older Adults
Published on: January 26, 2019
7.8K
Adversarial learning for semi-supervised pediatric sleep staging with single-EEG channel
Yamei Li1, Caijing Peng2, Yinkai Zhang3
1College of Electronic and Information Engineering, Southwest University, Chongqing, China.
Methods (San Diego, Calif.)
|April 1, 2022
Summary
This study introduces Bi-Stream Adversarial Learning network (BiSALnet) for more accurate pediatric sleep staging. The novel method improves semi-supervised learning by generating high-confidence pseudo-labels, addressing class imbalance for better sleep classification.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Medicine
- Sleep Medicine
Background:
- Pediatric sleep staging presents unique challenges due to complex sleep structures, differing from adult patterns.
- Automatic sleep staging aids clinical diagnosis but faces limitations with limited labeled data.
- Semi-supervised learning offers a solution by utilizing both labeled and unlabeled data, reducing annotation burden.
Purpose of the Study:
- To develop an advanced semi-supervised learning method for accurate pediatric sleep staging.
- To address the class-imbalance problem in sleep staging that hinders traditional semi-supervised approaches.
- To enhance the confidence and quality of pseudo-labels generated for network optimization.
Main Methods:
- Proposed a novel Bi-Stream Adversarial Learning network (BiSALnet) employing adversarial learning in Student and Teacher branches.
- Implemented a similarity measurement function to minimize output divergence and a discriminator to improve discriminative ability.
- Integrated a symmetric positive definite (SPD) manifold structure and an attention feature fusion module for robust feature extraction and classification.
Main Results:
- Achieved an overall classification accuracy of 0.80, kappa of 0.73, and F1-score of 0.76 on a pediatric dataset.
- Demonstrated strong performance on the public Sleep-EDF dataset with accuracy of 0.91, kappa of 0.85, and F1-score of 0.77.
- Showcased comparable results to state-of-the-art supervised methods using limited labeled data.
Conclusions:
- BiSALnet effectively generates high-confidence pseudo-labels, significantly improving semi-supervised pediatric sleep staging.
- The method demonstrates robustness and generalizability across different datasets, including pediatric and public sleep datasets.
- BiSALnet offers a promising approach for accurate automatic sleep staging with reduced reliance on extensive physician annotations.
