AUTHOR=Feng Li-Xiao , Li Xin , Wang Hong-Yu , Zheng Wen-Yin , Zhang Yong-Qing , Gao Dong-Rui , Wang Man-Qing 

TITLE=Automatic Sleep Staging Algorithm Based on Time Attention Mechanism

JOURNAL=Frontiers in Human Neuroscience

VOLUME=Volume 15 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2021.692054

DOI=10.3389/fnhum.2021.692054

ISSN=1662-5161

ABSTRACT=The most important part of sleep quality assessment is the automatic classification of sleep stages. Sleep staging is helpful in the diagnosis of sleep-related diseases. This study proposes an automatic sleep staging algorithm based on the time attention mechanism. Time-frequency and non-linear features are extracted from the physiological signals of six channels and then normalized. The time attention mechanism combined with the two-way bi-directional gated recurrent unit (GRU) is used to reduce computing resources and time costs, and the conditional random field (CRF) is used to obtain information between tags. After five-fold cross-validation, the recognition accuracy of the five states in the sleep phase has reached a robust 0.88. The Wake, N2, and REM periods can reach an accuracy rate of more than 0.9. The overall accuracy (ACC), F1 score (F1), Kappa coefficient (Kappa), and N1 sensitivity (SEN-N1) obtained by the proposed model are 0.908, 0.91, 0.877, and 0.61, respectively. In the study of sleep staging, the recognition rate of the N1 stage is low, and the imbalance has always been a problem. Therefore, this study introduces a type of balancing strategy. By adopting the proposed strategy, SEN-N1 and ACC of 0.7 and 0.86, respectively, can be achieved. The experimental results show that compared to the latest method, the proposed model can achieve significantly better performance and significantly improve the recognition rate of the N1 period. The performance comparison of different channels shows that even when the EEG channel is not used, considerable accuracy can be obtained.