Exploring Time Domain Features And Cognitive Mechanism Of Sleep EEG In OSAS Patients Based On Deep Convolutional Neural Network

Sleep is closely related to psychological and cognitive functions.Sleep disorders will lead to mental illness and cognitive dysfunction.Due to the limitation of experimental paradigm and data processing techniques,past studies only focused on the brain processing mechanisms of different sleep stages,unable to identify specific neural activity patterns and their effects on psychology and cognition.With machine learning,it is expected to analyze the relationship between sleep process and brain activity pattern more detailedly.Obstructive Sleep Apnea/hypopnea Syndrome(OSAS)as a sleep disorder with high incidence rate,it will damage patients’ attention,executive function,and memory.This study analyzes the brain mechanism of cognitive and emotional disorders based on big data.We first propose and implement a new method based on deep convolutional neural networks,which can extract time-domain features of sleep data lacking time alignment and of unequal length,in order to make a more microscopic analysis of sleep.Secondly,we analyze the macro-and microscopic sleep features between OSAS patients and healthy people and patients with other sleep disorders in multiple datasets,and try to use the differential interpretation of OSAS patients and healthy people to OSAS cognitive and emotional disorders.The results of macro-feature analysis showed that the proportion of light sleep in OSAS patients did not increase.In contrary,the proportion of deep sleep(stage N2 and N3)in OSAS patient was higher than that in healthy people,especially the N2 period sleep characterized by sleep spindle.Besides,we found that the OSAS features were progressive with the severity.By comparing to other types of sleep disorders,we found that only macroscopic features cannot explain unique form of cognitive impairment of OSAS patients.After that,we analyzed the microscopic time domain features using pre-training network and transfer learning,and visualization method.Results showed that time domain EEG of OSAS patients had more significant differences compare to healthy people in SWS and REM stage.These differences have certain frequency domain features,which show that the difference waveform of OSAS patients has more slow wave components,and the amplitude fluctuation of OSAS patients is greater in the 25 Hz oscillation of REM sleep period.With machine learning,this study provides a time domain feature extraction method with high time resolution and analyzes the specific patterns of EEG differences between OSAS patients and normal population.By means of tACS,rTMS and other experimental methods,the time domain features of a certain sleep stage will be changed,and the cognitive effects will be investigated to achieve effective intervention.