Key result
Semi-supervised CRFNet detects OSA from single-lead ECGs with ~90% accuracy using minimal labeled data.
Why the study?
Deep learning models for automated obstructive sleep apnea detection from single-lead ECG signals require a large amount of manually labeled data.
Does an enhanced semi-supervised learning method combined with a hybrid CNN-ResNet-Transformer architecture (CRFNet) using 20% labeled data achieve comparable Obstructive Sleep Apnea detection accuracy to fully supervised learning on single-lead ECG datasets?
Does an enhanced semi-supervised learning method combined with a hybrid CNN-ResNet-Transformer architecture (CRFNet) using 20% labeled data achieve comparable Obstructive Sleep Apnea detection accuracy to fully supervised learning on single-lead ECG datasets?
A semi-supervised deep learning framework combining CNN, ResNet, and Transformer modules enables accurate automated detection of obstructive sleep apnea from single-lead ECG signals using only 20% labeled data, matching fully supervised performance.
Answers are grounded in this paper and cite their sources.
May facilitate scalable OSA screening from single-lead ECG; leaves open prospective validation before clinical adoption.
Obstructive Sleep Apnea (OSA) is a common sleep disorder that poses significant health risks. Given the close relationship between cardiac and respiratory functions, respiratory changes caused by OSA can be effectively captured through electrocardiogram (ECG) signals. This study proposes an OSA detection algorithm based on single-lead ECG signals combined with simplified denoising processing. Specifically, to address the high demand for manually labeled data in deep learning, we propose an enhanced and easy-to-implement semi-supervised learning method (ESSL) to explore accurate OSA detection with a small amount of manually labeled data. A hybrid architecture (CRFNet) integrating Convolutional Neural Network (CNN), ResNet, and multi-scale Transformer modules is also proposed. On the PhysioNet Apnea-ECG dataset, the semi-supervised learning method combined with CRFNet achieved an accuracy of 89.82±0.39%, specificity of 88.94±1.04%, and sensitivity of 90.74±0.84% using 20% labeled data, which is close to the fully supervised learning results with 100% labeled data. On the Huashan Hospital dataset (HHD), the same model using the semi-supervised learning method also approaches the results of fully supervised learning. In the intra-dataset evaluation on another public dataset (UCDDB), the proposed ESSL similarly achieves near-fully supervised performance, while in the cross-dataset evaluation, the ESSL method improves the cross-dataset generalization of automatic OSA detection. Comparative results demonstrate competitive classification performance, while the intra- and cross-dataset evaluations further support the robustness and generalization capability of the proposed framework. By relying on easily accessible single-lead ECG signals and a semi-supervised learning approach that reduces the need for extensive labeling, this method provides a cost-effective and accurate solution for automated OSA detection.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied Obstructive Sleep Apnea (OSA). Enhanced semi-supervised learning (ESSL) with hybrid CNN-ResNet-Transformer (CRFNet) using 20% labeled single-lead ECG data vs. Fully supervised learning with 100% labeled data was evaluated on OSA detection accuracy on the PhysioNet Apnea-ECG dataset. Enhanced semi-supervised learning with CRFNet using 20% labeled single-lead ECG data achieved 89.82% accuracy, 90.74% sensitivity, and 88.94% specificity for OSA detection.
Synapse has enriched 10 related papers. Consider them for comparative context: