Research on Sleep Stage Classification of Electroencephalogram Signals Based on CNN and LSTM

Abstract

Chronic sleep deprivation can seriously affect physical and mental health and increase the risk of many chronic diseases. Sleep staging methods can assess sleep quality, but the manual interpretation is inconvenient and subjective. Establishing efficient automated sleep staging models is crucial to saving time and cost, improving efficiency, and helping patients with sleep disorders receive timely treatment. In recent years, artificial intelligence has been rapidly developing in the medical field, diagnostic efficiency has become an auxiliary tool, and even the accuracy rate exceeds that of doctors. In this paper, deep learning technology is applied to sleep EEG signal staging, exploring the basic theory and preprocessing method of EEG signal, and comparing the performance of the automatic sleep staging model through experiments. The process mainly involves processing the original dataset, establishing a standard dataset, and extracting features using a band-pass filter combined with VMD and FFT methods. Two automatic sleep staging models based on LSTM and LSTM+CNN were proposed. The models, which integrate VMD and FFT methods, achieved accuracies of 87% and 92%, respectively. Experimental results showed that the LSTM+CNN model, which combines VMD and FFT methods, performed better in terms of classification accuracy and loss values. Compared to using LSTM alone for sleep staging, it demonstrated outstanding classification performance. This technology is expected to provide healthcare institutions and doctors with faster and more accurate sleep monitoring and diagnostic tools to improve patients’ sleep health and quality of life.