Automated sleep staging model for older adults based on CWT and deep learning.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Manual sleep staging is time-consuming and subjective, impacting sleep disorder diagnosis.
• Automatic sleep staging models lack age-specific considerations, despite age-related sleep architecture changes.
• Older adults are more susceptible to sleep disorders due to altered sleep patterns, necessitating specialized models.

Purpose of the Study:

• To develop an accurate automatic sleep staging model tailored for older adults.
• To address the challenges posed by age-related alterations in sleep architecture.
• To improve the efficiency and reliability of sleep disorder diagnosis in the elderly population.

Main Methods:

• Established a 3D time-frequency feature fusion map dataset using continuous wavelet transform.
• Developed RICM-SleepNet, incorporating Inception modules, CBAM attention, and multiscale feature concatenation.
• Determined optimal channel signals from the Sleep-EDF expanded dataset.

Main Results:

• RICM-SleepNet achieved 87.66% accuracy and a κ value of 0.8492 on the specialized dataset.
• The model outperformed established baseline models including GoogLeNet, ResNet50, and DenseNet121.
• Statistical analysis (Kruskal-Willis test) confirmed the model's superior performance (p=0.0014).

Conclusions:

• RICM-SleepNet demonstrates significant potential for accurate automatic sleep staging in older adults.
• The model's performance surpasses existing methods, offering a promising tool for clinical applications.
• Further validation is recommended for broader implementation in managing age-related sleep issues.