Application of convolutional neural network-based biosensor and electroencephalogram signal in sleep staging

Abstract

Currently, sleep disorders have become a risk factor for various diseases; to analyze sleep disorders, sleep staging is a useful tool. The convolutional neural network (CNN) is improved by integrating with fine-grained segmentation and fuzzy neural network (FNN). Then the improved CNN model is employed for automatic sleep staging of electroencephalogram (EEG) signals. Data combinations of different sleep signals in the Sleep-EDFx sleep database of PhysioNet, including single-channel EEG, EEG + electrooculogram (EOG), EEG + electromyography (EMG), and EEG + EOG + EMG, are chosen and respectively preprocessed. The training set, validation set, and test set are selected randomly. Then, these datasets are input into the deep learning classification model. The sleep states are staged through the deep learning CNN model, and the staging results are output. The staging accuracy (Acc) and the macro-F1 (F1) score are utilized to evaluate the model’s performance, and the model results are compared with the expert staging results. The multichannel EEG + EOG + EMG model has the best effect; its ACC is 91.23%, and F1 is 86.35%. The Pearson correlation coefficient between model staging results and expert staging results is 0.86. The multichannel EEG + EOG + EMG sleep staging CNN model’s performance is better, with high staging accuracy, low signal-to-noise ratio, and short calculation time. The improved CNN model’s sleep features are highly consistent with those evaluated by experts, providing a practical basis for sleep staging. The results are expected to present a theoretical basis for clinical sleep staging.