CausalAttenNet: A Fast and Long-Term-Temporal Network for Automatic Sleep Staging With Single-Channel EEG

Automatic sleep staging is of great significance for the diagnosis and treatment of sleep disorders. However, the published works for sleep staging have not efficiently characterized temporal information from electroencephalogram (EEG), and thus face the problems of future information leakage and high time cost for model training. To address the above issues, this article proposes a novel network architecture named CausalAttenNet. It avoids future information leakage and extracts time-relevant sleeping features with a wide temporal receptive field through causal dilation convolution. The traditional recursive network is replaced by multihead attention (MHA) to largely reduce time cost and form a faster network for sleep staging. The accuracy rates of CausalAttenNet on the Sleep-EDF-20, Sleep-EDF-78, and Sleep Heart Health Study (SHHS) datasets are 84.7%, 81.6%, and 83.3%, respectively, which are better than those of the state-of-the-art methods. The performances of the proposed CausalAttenNet on different datasets demonstrate its effectiveness and superiority in the field on sleep staging. Individual difference is allowed and hand-crafted feature design is unnecessary for CausalAttenNet. The high accuracy and efficiency of the proposed method indicate its clinical potential for detecting sleep disorders.