Moving from Manual to Automated Sleep Staging with Federated Learning.

Area of Science:

• Artificial Intelligence
• Sleep Medicine
• Machine Learning

Background:

• Manual sleep stage scoring is time-consuming and subjective.
• Sharing sensitive patient data for AI model training raises privacy concerns.
• Automated sleep analysis is crucial for diagnosing sleep disorders.

Purpose of the Study:

• To explore Federated Learning (FL) for automated sleep stage classification.
• To develop a privacy-compliant AI model using distributed patient data.
• To evaluate the performance of FL in sleep stage classification.

Main Methods:

• Federated Learning (FL) was implemented on the ODIN platform.
• An AI model was trained on distributed patient data (148 sleep recordings, 117 patients).
• Model performance was validated against manual scoring and centralized training.

Main Results:

• The FL model achieved an accuracy of 74.2% for sleep stage classification.
• This accuracy is comparable to traditional manual scoring methods.
• The FL approach demonstrated scalability and privacy compliance.

Conclusions:

• Federated Learning offers a scalable and privacy-preserving solution for automated sleep stage classification.
• This framework can reduce clinician workload and facilitate multi-institutional collaboration.
• The developed system is suitable for immediate clinical application in sleep disorder diagnosis.