CareSleepNet: A Hybrid Deep Learning Network for Automatic Sleep Staging.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Automatic sleep staging from Polysomnography (PSG) is crucial for sleep assessment and disease diagnosis.
• Existing methods often overlook global features within sleep epochs and cross-modality relationships between EEG and EOG signals.

Purpose of the Study:

• To propose CareSleepNet, a novel hybrid deep learning network for advanced automatic sleep staging.
• To address limitations in existing sleep staging techniques by incorporating global features and cross-modality context.

Main Methods:

• Developed a multi-scale Convolutional-Transformer Epoch Encoder for local and global feature extraction.
• Implemented a Cross-Modality Context Encoder using a co-attention mechanism to model inter-signal relationships.
• Utilized a Transformer-based Sequence Encoder for capturing temporal dependencies across sleep epochs.

Main Results:

• CareSleepNet achieved state-of-the-art performance on three datasets: SSND, Sleep-EDF-153, and ISRUC.
• Ablation studies and attention visualizations confirmed the effectiveness of individual modules and modality contributions.

Conclusions:

• The proposed CareSleepNet effectively integrates local, global, and cross-modality features for superior automatic sleep staging.
• This deep learning approach offers a promising advancement for objective sleep assessment and clinical diagnosis.