ConvTransSleepNet: A Novel Framework for EEG-based Automatic Sleep Staging in Wearable Sleep Sensors

Sleep stage classification is pivotal for the accurate diagnosis of sleep disorders. Traditional methodologies, characterized by their labor-intensive nature, often fall short in precision, particularly when applied to single-channel wearable technologies. Introducing ConvTransSleepNet, a novel framework that synergizes the capabilities of convolutional neural networks with transformer layers, we mark a significant leap forward in accuracy for single-channel EEG-based sleep stage determination. Upon evaluation using the noise-reduced Fpz-Cz EEG signals from the Sleep-EDF Expanded dataset, our model demonstrated superior performance, achieving an average accuracy of 0.84 and an F1 score of 0.79 across 78 subjects, thereby surpassing competing algorithms. ConvTransSleepNet's efficiency in rendering precise sleep stage assessments underscores its potential for integration into an array of wearable devices, facilitating broader access to advanced sleep analysis.