Federated Learning Promises Better Sleep Medicine Assistance while Preserving Data Privacy

Researchers at the Dalian University of Technology have published a study on the application of federated learning in sleep medicine assistance, which has shown promising results in improving model performance while preserving data privacy. The study utilized a distributed data-privacy preserving federated learning method to develop a CNN-Transformer model for automatic sleep staging, achieving better overall and N1 classification performance compared to traditional local learning and data-centralized learning.

Key Takeaways:

• The study demonstrates the effectiveness of federated learning in addressing data privacy restrictions when collecting data from multiple sources to develop classification models.
• The proposed method, which uses a CNN-Transformer model and an anti-class ratio (ACR) weighted loss function, achieves better classification performance compared to traditional local learning and data-centralized learning.
• The study evaluates the model using the Sleep-EDF database and achieves better overall and N1 classification performance.
• The use of federated learning in sleep medicine assistance shows promise in improving model performance and preserving data privacy, particularly when dealing with insufficient data.
• The proposed method is a promising alternative for multiple clients developing sleep staging algorithms, particularly when dealing with data privacy restrictions.
• The study was conducted by researchers from the Dalian University of Technology, Faculty of Medicine, School of Biomedical Engineering.
• The researchers involved in the study include Hang Liu, Hongjin Li, Ziyi Wang, Yukai Cai, Chuanshuai Yang, Fengyu Cong, and Xiaohui Zhao.
• The study was published in the journal Biomedical Signal Processing and Control in 2025.

Statistics:

• The proposed method achieves better overall classification performance compared to traditional local learning and data-centralized learning.
• The study evaluates the model using the Sleep-EDF database and achieves better N1 classification performance.
• The results suggest that federated learning outperforms local learning approach in terms of classification performance.
• The study compares the experimental results with three independent heterogeneous public sleep databases.

Sources:

• NewsRx. Researchers from Dalian University of Technology Detail Findings in Information and Data Privacy (Distributed Data-privacy Preserving Federated Learning Method for Sleep Classification). Information Technology Newsweekly. 2025; p 739.
• Distributed Data-privacy Preserving Federated Learning Method for Sleep Classification. Biomedical Signal Processing and Control, 2025;109.