Research And Application Of Deep Learning Based Sleep Staging Recognition Algorithm

Sleep is a physiological phenomenon,and the quality of sleep affects everyone’s life.According to the World Health Organization,27% of the world’s population has sleep problems,and the National Health and Wellness Commission has included the current rate of insomnia and the average daily sleep time of adults into the indicators of the Healthy China Initiative,and the promotion of sleep health has become an important element of the Healthy China Initiative.The diagnosis and treatment of modern clinical sleep medicine in China has been carried out for more than 40 years.However,the polysomnographic monitoring technology is far from meeting the realistic clinical sleep disease diagnosis and treatment work requirements.Polysomnography(PSG),also known as sleep electroencephalography,is the primary measurement and diagnostic tool in sleep centers.The definition of sleep is based on behavioral and physiological criteria that classify it into two states: non-rapid eye movement sleep as well as rapid eye movement sleep.Non-Rapid Eye Movement Sleep(NREM Sleep)is divided into three stages(N1,N2,N3)and Rapid Eye Movement Sleep(REM Sleep)is characterized by rapid eye movements,muscle dystonia and electroencephalogram(EEG)asynchrony.Sleep stages are an important indicator for diagnosing sleep status,and traditional manual identification methods are inefficient and not guaranteed in quality.Therefore,it is necessary to implement sleep staging algorithms.On this basis,it is important to automate a large number of simple and repetitive tasks by supporting application software to reduce the threshold of sleep monitoring technology use and thus complete sleep-assisted diagnosis.In this context,a new recognition algorithm based on deep neural network is proposed and a software system is designed and implemented to run with the recognition algorithm.The main work of this paper is as follows.To address the problem of poor accuracy of sleep staging recognition algorithm N1 stage recognition,this paper proposes the Gated Attentional Symmetric Neural Network(GAS-NN)algorithm.The algorithm introduces a channel attention mechanism and channel-level gated recurrent units to build a convolutional neural network,which uses EEG-EOG bimodal physiological electrical signals to complete end-to-end sleep staging recognition.It was experimentally validated that GAS-NN achieved an average 1- of 79.2% on the sedf＿sc dataset and an 1-of 60.6% for the most difficult to identify N1 stage of sleep staging.GAS-NN achieves an average 1- of 80.2% on the sedf＿st dataset and 82.3% on the DCSM dataset.The GAS-NN algorithm starts from the physiological characteristics,temporal characteristics and waveform characteristics of sleep physiological electrical signals,makes full use of the time-dependent and physiological characteristics of sleep physiological electrical signals,introduces Electroocular information to assist in determining the N1 stage,the model achieves high-precision recognition,takes into account the computational efficiency and generalization performance,completes computer-aided sleep staging,and lays the foundation for computer-aided sleep quality assessment.In order to reduce the threshold of using sleep staging recognition algorithm and eliminate a lot of simple and repetitive labor,this paper constructs an application software system based on microservice technology,which completes automated sleep physiological electrical signal data transmission,neural network algorithm driving,algorithm recognition result feedback,and physiological electrical signal acquisition experimental data tracking.This paper implements a system runtime environment design based on containerization technology and provides an extensible interface to access other physiological signal recognition algorithms to facilitate system function expansion and horizontal capacity expansion.In addition,high-performance distributed middleware is introduced and memory-level optimization is completed to ensure efficient and stable operation of the system.This paper stress tests the software system,and the test results show that 95% Line of the core interface of the application software in this paper is about 100 ms,and the cross-service call level does not exceed 3 layers,which avoids the service timeout and service unavailability caused by complex calls.The application software automates a large amount of work before and after the operation of the sleep staging recognition algorithm,providing an automated sleep staging solution from the acquisition of physiological electrical signals to the feedback of calculation results.The system is efficient,reliable,stable and easy to use,lowering the application threshold of the sleep staging recognition algorithm and laying the foundation for tracking and monitoring sleep quality.