Automatic Sleep Staging Method Using EEG Based on STFT and Residual Network

Sleep is a vital physiological process that affects both physical and mental health, with sleep disorders linked to various conditions such as mental illnesses and cardiovascular diseases. Accurate sleep stage classification is crucial for assessing sleep quality and diagnosing sleep disorders; however, traditional manual sleep staging methods are time-consuming and prone to human bias. This study introduces an enhanced method for automatic sleep staging that addresses the challenges of low training efficiency and reliance on extensive labeled datasets. Initially, high-dimensional time-frequency features of EEG signals are extracted using a short-time Fourier transform (STFT), transforming unidimensional signal data into multidimensional image datasets. Subsequently, the deep-seated temporal and spectral features of sleep stages are unearthed and learned through an improved ResNet-18 residual network, facilitating the automated classification of sleep stages. The proposed method demonstrated a classification accuracy of 85.71% on seven selected full-night recordings from the Sleep-EDF dataset, with a macro-average F1-score of 85.05%. Comparisons with other methods confirm the effectiveness of the proposed method in improving sleep quality evaluation and diagnosis. The results indicate the potential for using this approach in real-time clinical applications, offering a reliable and efficient solution for sleep disorder diagnosis. This work lays a foundation for further exploration of automated systems in sleep medicine, contributing to more accurate and accessible sleep health assessments.