Enhancing EEG-based sleep staging efficiency with minimal channels through adversarial domain adaptation and active...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate sleep-stage classification is vital for sleep research and healthcare.
• Traditional deep learning and domain adaptation methods face challenges with limited labeled data and subtle class distinctions.

Purpose of the Study:

• To develop a novel framework, adversarial domain adaptation with active deep learning (ADAADL), to enhance sleep-stage classification accuracy.
• To effectively leverage unlabeled data and reduce annotation costs in sleep studies.

Main Methods:

• The ADAADL framework combines adversarial learning with active learning (AL) strategies.
• Utilizes two sleep-stage classifiers as discriminators for refined feature alignment.
• Incorporates entropy measures and cross-entropy loss for better utilization of unlabeled data.
• Employs an active learning component to iteratively select and label informative data points.

Main Results:

• ADAADL generates robust and transferable features for sleep-stage classification.
• Demonstrates significantly improved classification accuracy compared to existing domain adaptation methods on benchmark EEG datasets.
• Shows enhanced generalization to unseen data through active learning.

Conclusions:

• The ADAADL framework represents a significant advancement in sleep-stage classification.
• Offers a scalable and accurate solution for real-world sleep monitoring applications.
• Contributes to a deeper understanding of sleep dynamics by improving sleep stage modeling.