Assessing Obstructive Sleep Apnea Severity During Wakefulness via Tracheal Breathing Sound Analysis

Area of Science:

• Respiratory Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Obstructive sleep apnea (OSA) is underdiagnosed and linked to increased accident risk and perioperative complications.
• The gold standard diagnostic tool, polysomnography (PSG), is expensive, time-consuming, and not widely accessible.
• Current diagnostic methods for OSA severity lack speed and accessibility.

Purpose of the Study:

• To develop a fast, objective, and non-invasive method for detecting obstructive sleep apnea severity.
• To analyze tracheal breathing sounds (TBS) recorded during wakefulness for OSA classification.
• To evaluate machine learning models for accurate OSA severity assessment.

Main Methods:

• Collected tracheal breathing sounds (TBS) from 199 subjects across Non-OSA, Mild, Moderate, and Severe categories.
• Extracted features from TBS and combined with anthropometric data for classification.
• Trained and blind-tested machine learning models, including Support Vector Machine and Random Forests, on shuffled datasets.

Main Results:

• The Support Vector Machine model for Non-OSA vs. Severe-OSA achieved 88.2% accuracy, 83.3% sensitivity, and 90.9% specificity.
• The Random Forests model for Non-OSA vs. Mild-OSA demonstrated 100% sensitivity but 81.2% accuracy.
• The proposed method demonstrated reliable and robust performance for OSA severity screening.

Conclusions:

• Tracheal breathing sound analysis offers a promising, rapid, and non-invasive approach for obstructive sleep apnea severity screening.
• This method can potentially be implemented in clinical settings for quick OSA assessment in under 10 minutes.
• The findings support the development of accessible tools for managing OSA and reducing associated health risks.