Exploiting temporal and nonstationary features in breathing sound analysis for multiple obstructive sleep apnea...

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Signal Processing

Background:

• Polysomnography (PSG) is the standard for obstructive sleep apnea (OSA) diagnosis but is costly and inconvenient.
• There is a need for accessible, repetitive screening methods for OSA risk assessment.
• Breathing sounds offer a potential data source for non-invasive OSA evaluation.

Purpose of the Study:

• To develop a novel four-class classification model for obstructive sleep apnea (OSA) severity.
• To utilize patient breathing sounds as the primary data input for OSA classification.
• To create an affordable and convenient alternative for OSA screening.

Main Methods:

• Breathing sounds were collected from 83 subjects undergoing PSG.
• Two novel signal processing techniques were applied: transition probability of sound energy and properties of dimension-reduced cyclic spectral density.
• Feature selection and support vector machines were used for model training and evaluation via leave-one-out cross-validation.

Main Results:

• The proposed model achieved 79.52% accuracy for four-class OSA severity classification.
• For OSA detection (Apnea-Hypopnea Index threshold 5), the model showed 98.0% sensitivity and 75.0% specificity.
• The classification model outperformed existing methods using breathing sounds for OSA severity.

Conclusions:

• Breathing sound analysis provides a viable method for obstructive sleep apnea (OSA) screening.
• The developed model can offer detailed OSA severity information non-invasively.
• This approach can serve as a valuable component of an OSA screening test, enhancing accessibility and patient convenience.