A Fine-Grained Diagnostic Method for Sleep Apnea Based on Target Detection Network With SpO2 and Nasal Airflow

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Sleep Medicine

Background:

• Current sleep apnea diagnostic tools prioritize apnea-hypopnea index (AHI) estimation and severity classification.
• Existing methods often overlook precise annotation of abnormal respiratory events.

Purpose of the Study:

• To develop a fine-grained diagnostic method for accurate sleep apnea event detection and clinical indicator assessment.
• To introduce the apnea-hypopnea event detector (AHE-Detector) for enhanced sleep apnea diagnosis.

Main Methods:

• Proposed a novel AHE-Detector model based on target detection principles.
• Fused time-frequency features from pulse oxygen saturation and nasal airflow signals.
• Employed a three-module architecture: feature fusion, multi-scale feature extraction, and event prediction.

Main Results:

• The AHE-Detector accurately identifies abnormal respiratory events on a continuous time axis.
• Achieved considerable performance in detecting events, distinguishing hypopnea and apnea, and classifying apnea subtypes.
• Enabled accurate simultaneous estimation of AHI and apnea-hypopnea event duration index.

Conclusions:

• The AHE-Detector represents a significant advancement in sleep apnea diagnostics.
• This method is the first to accurately and simultaneously estimate two critical sleep apnea clinical indicators.
• Outperformed existing methods in event detection, clinical index estimation, and severity classification.