A Human-Computer Collaborative Sleep Scoring System for Clinical Needs

This paper addresses the time-consuming sleep scoring process performed by technicians in overnight polysomnography (PSG), which influences subsequent treatment decisions by physicians. To alleviate this issue, an automatic sleep scoring system is proposed. Unlike previous approaches, this paper focuses on clinical relevance. Collaborating with the Sleep Medicine Center of National Cheng Kung University Hospital, a user-friendly interface akin to the existing system is established. This interface supports automatic scoring of sleep stages (Wake, N1, N2, N3, and REM), and respiratory events (Apnea and Hypopnea). The results are both adjustable and interpretable on the interface. Evaluation showcases promising results, with overall sleep stage agreement at 76.5%, apnea recall at 0.86 and precision at 0.76, and hypopnea recall at 0.62 and precision at 0.7. Notably, the system's performance aligns with technician-scored results using workshop data, demonstrating robustness through clinical application and time-saving efficiency. This paper underscores the potential for practical clinical implementation in assisting sleep technicians.