Point out the mistakes: An HMM-based anomaly detection algorithm for sleep stage classification

Accurate sleep stage scoring is essential for diagnosing sleep disorders. Current automated sleep staging methods often exhibit staging errors, which can be interpreted as anomalies. Detecting these anomalies is crucial for improving staging accuracy. Most existing approaches modify staging based on predefined conditions but lack effective methods for localizing and identifying anomalies. In this study, we propose an anomaly detection method utilizing the Hidden Markov Model (HMM), a time-series modeling technique, to detect anomalies in sleep staging results. Evaluating our approach with four classical models as pre-classifiers, we achieve anomaly detection precisions of 0.760, 0.577, 0.631, and 0.613. Assuming that all detected anomalies are corrected, the pseudo-accuracies improve to 0.964, 0.929, 0.950, and 0.929, respectively. Our results indicate that the proposed method significantly enhances stage recognition accuracy, especially for stage N1, which is critical for diagnosing sleep-related disorders. Notably, approximately 28.6% of epochs require reinterpretation by sleep technicians to achieve these improvements.