Modality-Specific Feature Selection, Data Augmentation and Temporal Context for Improved Performance in Sleep...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Accurate sleep staging is crucial for diagnosing sleep disorders and understanding sleep physiology.
• Existing sleep staging systems often struggle with cross-dataset generalizability and performance on clinical populations.

Purpose of the Study:

• To develop an effective sleep staging system that performs well on diverse datasets, including healthy and clinical populations.
• To achieve high accuracy in cross-dataset sleep staging experiments.

Main Methods:

• A hierarchical model using multiple binary classifiers was designed.
• Electroencephalography (EEG), electrooculography (EOG), and electromyography (EMG) signals were utilized.
• Feature selection, random undersampling, boosting, temporal context, and data augmentation were employed to enhance performance.

Main Results:

• The best performance was achieved using features from EEG, EMG, and EOG signals at each hierarchical level.
• Average accuracies ranged from 73.7% to 90.0% across multiple datasets (Sleep-EDF, Exp Sleep-EDF, ISRUC-S1, S2 and S3, DRMS-SUB, DRMS-PAT).
• The proposed method outperformed state-of-the-art approaches on most datasets and demonstrated robust cross-dataset performance exceeding 80%.

Conclusions:

• The proposed hierarchical sleep staging system demonstrates superior performance and generalizability compared to existing methods.
• The integration of multi-modal signals (EEG, EOG, EMG) and advanced techniques significantly improves sleep staging accuracy.
• This system holds promise for clinical applications and advancing sleep research.