Design Of Multimodal Intelligent Sleep Monitoring System

Sleep disorders are prevalent health problems that can impact brain function,cognitive abilities,and even pose life-threatening risks.Currently,the diagnosis of sleep disorders requires the use of expensive polysomnography(PSG)equipment in hospitals,followed by manual analysis conducted by experts.This approach is not suitable for large-scale applications.Furthermore,the accuracy of simplified monitoring devices like wristbands falls short of clinical standards.Therefore,the development of a portable and home-use medical-grade sleep monitoring system holds significant value for home sleep monitoring.In this paper,we present a multimodal sleep monitoring system based on Android and a cloud service platform.The system utilizes data from electroencephalography(EEG),electrocardiography(ECG),respiration,and other modalities.We conduct research on feature extraction,feature selection,and sleep staging algorithms based on the collected data.(1)Addressing the limitations of conventional sleep monitoring devices,we design and implement a portable multimodal sleep monitoring system based on the STM32F40 x ZGT6 core and ADS1292 analog front-end.We develop a mobile application for Android platforms,enabling functionalities such as multimodal data acquisition,parsing,visualization,storage,upload,and labeling.Additionally,we establish a cloud service platform that integrates sleep monitoring algorithms and provides services such as report generation and data management.Through these designs,our system achieves portability,home-use capability,multimodality,and intelligence.(2)We synchronize the self-made system with PSG systems to record sleep data from14 healthy individuals.Sleep physicians annotate the sleep stages based on PSG data.We perform a multimodal analysis of the single-channel frontal electroencephalogram(EEG)and electrocardiogram(ECG)signals obtained by the self-made system,extracting a total of 71 features.In addition to the common EEG and ECG features,we propose cardiorespiratory and cardiocerebral coupling algorithms based on the Cardio Pulmonary Coupling(CPC)algorithm from Harvard Medical School.We extract 38 features to investigate the synchrony,directional connection,and mutual information between the cardiorespiratory and cardiocerebral systems.(3)To obtain optimal classifiers and feature combinations,we divide all features into five sets based on data modality.Using the Kruskal-Wallis multiple comparison test,we analyze the features and eliminate those with less significant differences.We then employ four classification algorithms for automatic sleep staging.The results demonstrate that the highest accuracy rates achieved by different classifiers were 64.29%,79.79%,84.32%,88.01%,and 89.09% for the traditional ECG feature set,ECG modal feature set,EEG modal feature set,EEG cardiopulmonary feature set,and EEG-ECG modal feature set,respectively.Furthermore,we apply a recursive feature elimination algorithm to remove redundant features,thus conserving computational resources on the cloud platform while maintaining accuracy.