Impact of Deep Learning Techniques on Automated Sleep Staging Diagnosis using BrainEEG Signals

Sleep, a fundamental physiological process, manifests as a cascade of changes in brain waves, muscle tension, eye movements, and heart and respiratory activities. Monitoring and analyzing sleep and sleep patterns is thus essential for understanding their impact on physical, mental, and emotional health. Traditional methods, exemplified by polysomnography (PSG) and Actigraphy, have long been the gold standards for sleep staging diagnosis but suffer from high costs, subjectivity in data interpretation and patient discomfort. Deep learning (DL), a subset of artificial intelligence, has lately emerged as a transformative force in sleep analysis. This paper comprehensively examines state-of-the-art DL models like Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) evaluating their exceptional potential to classify sleep stages accurately. The literature survey encompasses studies of various physiological signals through Electroencephalogram (EEG), Electrooculogram (EOG), Electromyogram (EMG), and Electrocardiogram (ECG) for automated sleep staging. Model performances are compared across multiple datasets like SleepEDF, SHHS, and PhysioNet, showeasing high classification accuracies-often exceeding 85%—when leveraging hybrid models with multimodal inputs. Key findings indicate that CNN and hybrid models offer high classification accuracy, particularly when multiple signals are utilized. Despite these advancements, challenges such as data quality, interpretability, and generalizability persist. The findings suggest that DL holds immense promise for improving diagnostic accuracy, scalability, and accessibility in sleep research. Furthermore, this review helps to identify open challenges and opportunities for future research, emphasizing the integration of personalized medicine and cloud-based healthcare solutions.