Abstract
(1) Background: Sleep Apnea Syndrome (SAS) poses a serious threat to human health. Existing SpO2-based automatic SAS detection models have a relatively low accuracy in detecting positive samples because they overlook the global information from the Apnea-Hypopnea Index (AHI). (2) Methods: To address this problem, we proposed a multi-task model for SAS detection and AHI prediction based on single-channel SpO2. Benefiting from the characteristics of the Broad Learning System (BLS), this model optimizes itself by leveraging the differences between all-night SpO2 information and sample SpO2 information, enabling the two tasks to promote each other. (3) Results: The model was verified using 7906 all-night SpO2 data from the publicly available Sleep Heart Health Study (SHHS) dataset, and the SAS detection performance has reached the state-of-the-art level. In addition, the performance of samples with different lengths in the two tasks was also explored. (4) Conclusions: The model we proposed can balance and effectively perform both SAS detection and AHI prediction simultaneously.