A novel approach to diagnose sleep apnea using enhanced frequency extraction network

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Signal Processing

Background:

• Sleep apnea-hypopnea syndrome (SAHS) is a prevalent respiratory disorder with serious health implications.
• Current polysomnography (PSG) diagnosis for SAHS is complex and expensive.
• Nasal airflow (NA) signals offer a direct and easily recordable measure of SAHS severity.

Purpose of the Study:

• To develop an automatic, low-cost, and efficient approach for detecting sleep apnea-hypopnea events using single-channel nasal airflow signals.
• To leverage wearable device technology for accessible SAHS monitoring.

Main Methods:

• An enhanced frequency extraction network was designed to factorize mixed feature maps by frequency.
• Spatial resolution of low-frequency components was reduced to optimize computational cost.
• Residual blocks, smaller filter groups with larger kernels, and spatial attention modules were employed for high-frequency component processing and feature extraction.

Main Results:

• The proposed network achieved outstanding performance in multiclass classification of SAH events.
• Achieved accuracy of 91.23%, sensitivity of 90.81%, and specificity of 90.59%.
• Demonstrated superior performance compared to state-of-the-art networks.

Conclusions:

• The developed approach offers a low-cost and high-efficiency solution for detecting SAH events.
• Shows significant potential for practical application in SAHS diagnosis and monitoring.
• Highlights the utility of wearable-based NA signal analysis for sleep disorder detection.