Sleep Stages Classification with Multi-modal Signals using Deep Learning

This study introduces a novel deep learning architecture for automated sleep stage classification addressing the limitations of traditional methods which are often laborintensive and subjective. Utilizing a hybrid Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) network. Our model analyzes multi-modal physiological data from the Sleep Multi-modal Electroencephalogram (SOMEG) dataset to accurately classify sleep stages. Our approach achieved a remarkable overall accuracy of 82.05% surpassing existing methods including Support Vector Machines and other published deep learning models that typically range from 79- 81% accuracy. Rigorous training, validation and testing demonstrate the robust generalization capabilities of our model. While confusion matrix analysis revealed areas for improvement in differentiating between stages with high interindividual variability specifically N1 and N2 in our findings highlight the significant potential of this architecture to advance automated sleep stage classification and contribute to more efficient and accurate diagnosis of sleep disorders.