Automatic sleep-stage classification based on residual unit and attention networks using directed transfer function of electroencephalogram signals

A Novel Continuous Sleep State Artificial Neural Network Model Based on Multi-Feature Fusion of Polysomnographic Data

“… 18 A new network for sleep staging called Deepsleepnet, 19 which utilized two convolutional neural networks to extract time-frequency features from EEG data, combined with Long Short-Term Memory (LSTM) networks to extract associations between different sleep stages, reached an accuracy rate of 82.0%. Some researchers developed a deep learning model based on images for automatic sleep staging, using Class Activation Maps to visualize key reasoning areas, 20 with an accuracy rate exceeding 80%, while others employed a multimodal architecture with residual units to address the vanishing problem in deep learning, 21 achieving a classification accuracy rate of 87.34% and an F1-score of 87.42%. For sleep staging based on deep learning, the requirements for feature extraction are not high; however, overall, the staging accuracy results do not significantly outperform machine learning algorithms based on feature selection.…”