Learning Unified Model for Sleep Health Monitoring

Sleep is a crucial physiological activity for humans, and accurately monitoring sleep stages and related disorders is essential for health diagnostics. This paper addresses the challenge of simultaneously predicting sleep staging, respiratory disorders, and oxygen saturation levels using a unified model based on a multitasking query transformer. The aim of this study is to improve the efficiency and accuracy of sleep health assessments by leveraging inter-task correlations. The proposed technique introduces a novel approach that enhances performance across multiple tasks while reducing the number of parameters compared to single-task models. Experiments conducted on the large-scale Sleep Heart Health Study (SHHS) dataset demonstrate strong performance, achieving an accuracy of 87.2% for sleep staging, a correlation coefficient of 0.398 for oxygen saturation prediction, an accuracy of 78.2% for respiratory disorder detection, and 58.67% for the classification of respiratory disorder severity. Further fine-tuning enhances task-specific performance, highlighting the model’s adaptability and effectiveness.