A Machine-Learning Model for Lung Age Forecasting by Analyzing Exhalations.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Pulmonology

Background:

• Spirometers are crucial for respiratory disease management but are typically hospital-based, limiting patient supervision.
• Digital alternatives are being developed, aiming for lower cost and wider accessibility.
• There is a need for user-friendly lung capacity metrics beyond traditional spirometry.

Purpose of the Study:

• To develop a machine learning method for estimating lung age from exhalation sound properties.
• To create a more accessible and user-friendly tool for lung health assessment.
• To popularize spirometry through novel, mobile-compatible metrics.

Main Methods:

• Utilized 188 exhalation sound samples from 91 males and 97 females (aged 17-67).
• Extracted 42 features, including spirometer and frequency-like characteristics, and gender.
• Applied traditional machine learning algorithms, including Quadratic Linear Discriminant analysis.

Main Results:

• The Quadratic Linear Discriminant algorithm achieved the highest accuracy (94.69%) when gender was not considered.
• Classification into 5-year lung age groups demonstrated high accuracy (94.69%), sensitivity (94.45%), and specificity (99.45%).
• Key features within exhalation audio were identified for accurate lung age classification.

Conclusions:

• The developed methodology reliably estimates lung age using machine learning on exhalation sounds.
• This approach offers a potential low-cost, mobile-device-compatible tool for early detection of lung abnormalities.
• The findings support broader accessibility and improved supervision for patients with respiratory diseases.