Toward Interpretable Sleep Stage Classification Using Cross-Modal Transformers.

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Sleep Medicine

Background:

• Accurate sleep stage classification is crucial for assessing sleep health.
• Deep learning models achieve human-level performance but suffer from black-box behavior, limiting clinical use.
• Existing methods lack interpretability and efficiency.

Purpose of the Study:

• To develop an interpretable deep learning model for sleep stage classification.
• To improve the efficiency of sleep staging algorithms in terms of parameters and training time.
• To provide a transparent alternative to current black-box deep learning models.

Main Methods:

• A novel cross-modal transformer architecture was developed for sleep stage classification.
• The model integrates a transformer encoder with a multi-scale 1D convolutional neural network for representation learning.
• Attention modules were utilized to enhance model interpretability.

Main Results:

• The proposed method achieved performance comparable to state-of-the-art sleep staging algorithms.
• The model demonstrated enhanced interpretability by leveraging attention mechanisms.
• Significant reductions in model parameters and training time were observed compared to existing methods.

Conclusions:

• The cross-modal transformer offers an interpretable and efficient solution for sleep stage classification.
• This approach addresses the limitations of black-box deep learning models in clinical settings.
• The method shows promise for advancing sleep health assessment through transparent AI.