Detection of Cheyne-Stokes Breathing using a transformer-based neural network.

Area of Science:

• Medical Technology
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Manual annotation of sleep disordered breathing (SDB), including Cheyne-Stokes Breathing (CSB), is resource-intensive.
• Accurate and efficient detection of CSB is crucial for patient management and research.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for automatic, sample-wise detection of CSB.
• To assess the algorithm's performance against certified sleep technicians.

Main Methods:

• Utilized 523 nocturnal polysomnographic (PSG) recordings from diverse sleep cohorts.
• Extracted 16 time-domain features and employed a transformer-inspired neural network.
• Applied post-processing to ensure physiologically relevant predictions.

Main Results:

• The algorithm achieved a F1-score of 0.76, comparable to certified sleep technicians.
• Performance in distinguishing CSB from severe obstructive sleep apnea was similar to human scorers.
• Demonstrated the feasibility of automated CSB detection.

Conclusions:

• The proposed deep learning algorithm shows significant promise for automated CSB detection.
• The model could serve as a valuable screening tool in clinical settings.
• Further validation is recommended to confirm its clinical utility.