AUTHOR=Zhuang Lan , Dai Minhui , Zhou Yi , Sun Lingyu 

TITLE=Intelligent automatic sleep staging model based on CNN and LSTM

JOURNAL=Frontiers in Public Health

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2022.946833

DOI=10.3389/fpubh.2022.946833

ISSN=2296-2565

ABSTRACT=In recent years, the study of sleep electroencephalogram automatic staging methods has important application value in the treatment and diagnosis of sleep disorders. Electroencephalogram (EEG) is a significant basis to treat and diagnose somnipathy. EEG is a non-linear, non-stationary signal which is very weak itself, and it needs accurate, efficient algorithms to extract feature information from the signal to be used in sleep stages. Conventional feature extraction methods have low efficiency and are difficult to meet the time-validity of fast staging; besides, due to insufficient priori knowledge, it can easily lead to omission of key features. Deep learning network such as convolutional neural networks (CNN) have robust capabilities of data analysis and mining. This paper introduces deep learning network into the research of sleep stages to fill the gap of traditional methods. There exists complementary information among different brain physiological signals. This paper adds fusion methods in the study and selects Long Short-term Memory (LSTM) as the classification network in order to improve the accuracy of sleep stages identification. Based on EEG and deep learning network, this paper proposes an automatic sleep stages method based on multi-channel EGG and uses CNN-LSTM to conduct the supervised learning of sleep stages on EEG and EOG samples. This network has 11 layers and the sleep data of every 30 seconds is divided into one stage. The predicted samples and the first 2 stages are taken as the input data. It doesn’t need any signal pre-processing or feature extraction, but it needs to do data augmentation (DA) on the data of unbalanced classes and delete special data and non-universal data. This paper uses MIT-BIH dataset to train and evaluate the models proposed and the experiment result shows that the EEG-based sleep stages method in this paper offers an effective method for the diagnosis and treatment of sleep disorders and has practical application value.