Deep learning for EEG-based sleep stage classification: a review.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Deep learning is increasingly used for sleep electroencephalogram (EEG) analysis, advancing sleep stage classification.
• This progress benefits sleep research and diagnosing neurological disorders.

Purpose of the Study:

• To systematically review deep learning models for sleep staging.
• To analyze architectural evolution, input strategies, preprocessing, datasets, and evaluation metrics.
• To identify challenges and propose future research directions.

Main Methods:

• Systematic review of deep learning models for sleep staging.
• Analysis of model architectures, input representations, preprocessing, datasets, and evaluation metrics.
• Cross-dataset accuracy and stage-wise F1 distribution analysis.

Main Results:

• Deep learning model performance is highly dataset-dependent.
• Current evaluation metrics (e.g., overall accuracy) fail to identify weaknesses on minority stages or pathological data.
• Key challenges include generalization, interpretability, and clinical applicability.

Conclusions:

• Future research should focus on diverse datasets, optimized architectures, enhanced interpretability, and clinical validation.
• The goal is to transition from algorithmic innovation to clinically reliable sleep staging.
• Addressing current limitations is crucial for advancing sleep research and neurological disorder diagnosis.