cVAN: A Novel Sleep Staging Method via Cross-View Alignment Network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Accurate sleep staging is crucial for assessing sleep quality and diagnosing sleep disorders.
• Current methods using multiple physiological signals show promise but neglect inter-view feature relationships at varying scales.
• A need exists for advanced methods that address feature scale alignment across different physiological signal data views.

Purpose of the Study:

• To propose a novel cross-view alignment network (cVAN) for improved sleep stage classification.
• To leverage scale-aware attention for adaptive alignment of features across different data views.
• To enhance the accuracy of sleep staging by effectively integrating multi-view physiological signal information.

Main Methods:

• Developed a novel cross-view alignment network (cVAN) incorporating residual-like and transformer-like sub-networks.
• Utilized spectral information from time-frequency images and temporal information from physiological signals.
• Implemented scale-aware attention to adaptively align learned feature scales across different data views by reorganizing feature maps.

Main Results:

• cVAN achieved state-of-the-art results in sleep stage classification on three public datasets.
• The proposed method demonstrated superior performance compared to existing sleep staging techniques.
• Scale-aware attention effectively aligned feature scales, enhancing classification accuracy.

Conclusions:

• The novel cVAN model significantly advances sleep stage classification accuracy.
• Cross-view alignment with scale-aware attention is effective for integrating multi-view physiological signals.
• This approach offers a promising direction for automated sleep quality evaluation and disorder diagnosis.