关键词: 睡眠分期, 睡眠呼吸暂停-低通气检测, 脑电图, 心电图, 深度学习, 多任务学习

Abstract: Sleep stage classification and sleep apnea-hypopnea detection are two key tasks for comprehensive assessment of sleep quality. However, the researchers implement the two tasks separately, without considering the intercorrelation between them. In fact, sleep stage and sleep apnea-hypopnea are associated. Sleep apnea-hypopnea syndrome is a kind of sleep fragmentation disorder which causes frequent changes in sleep stages and disrupts the normal sleep cycle. Meanwhile, sleep stage also affects the severity and duration of apnea and hypopnea. Considering the interrelationship, this paper proposes a multi-task framework for sleep staging and sleep apnea-hypopnea detection based on single-channel electroencephalogram (EEG) and single-lead electrocardiogram (ECG). The model utilizes a two-stream time-dependency learning module to extract the shared EEG and ECG features. An adaptive inter-task relevance learning module is used to capture the correlationship between sleep stage and sleep apnea-hypopnea and take advantage of the interaction to improve the performance. The performance of the proposed method was evaluated on two public datasets. The results show that the model can achieve the two tasks at the same time, with comparable performance on sleep staging and better performance of sleep apnea-hypopnea detection compared with the baseline methods. Especially, the proposed framework can detect more apneas and hypopneas, which is of great importance for primary screening of sleep apnea-hypopnea syndrome.

Key words: Keywords: sleep stage classification, sleep apnea-hypopnea detection, electroencephalogram, electrocardiogram, deep learning, multi-task learning