State-Dependent CNN-GRU Reinforcement Framework for Robust EEG-Based Sleep Stage Classification.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Machine Learning

Background:

• Automated analysis of biomedical signals aids sleep stage detection.
• Existing models struggle with imbalanced data and dynamic sleep states.

Purpose of the Study:

• Develop a robust algorithm for sleep state classification using electroencephalogram (EEG) data.
• Address challenges of imbalanced datasets and dynamic sleep states in current models.

Main Methods:

• Extracted dynamic, brain-inspired features (microstates, Lempel-Ziv complexity) from EEG.
• Developed a classifier using convolutional neural networks (CNN) and gated recurrent units (GRUs) within a reinforcement learning framework.
• Optimized feature set based on spectral ranges and classification performance.

Main Results:

• Achieved 98% classification accuracy using an optimized multivariate feature set.
• Demonstrated strong discriminative power of biomimetic features for sleep state classification.
• Framework utilized fewer EEG channels and reduced processing time compared to benchmarks.

Conclusions:

• Biomimetic principles in feature extraction and model design enhance automated sleep monitoring.
• The proposed framework shows potential for real-time deployment in sleep analysis.
• Findings support the development of novel diagnostic and therapeutic tools for sleep disorders.