Estimation of Static Lung Volumes and Capacities From Spirometry Using Machine Learning: Algorithm Development and...

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Data Science

Background:

• Spirometry is a common tool for diagnosing obstructive lung disease.
• Advanced lung function testing like body plethysmography is needed for detailed assessments but is not widely available.
• There is a need for methods to estimate static lung volumes using readily available spirometry data.

Purpose of the Study:

• To develop artificial intelligence (AI) algorithms for estimating lung volumes and capacities from spirometry measurements.
• To leverage machine learning techniques to extract clinically relevant information from spirometry data.

Main Methods:

• Utilized a large dataset of spirometry and lung volume measurements from the Mayo Clinic pulmonary function test database (2001-2022).
• Applied various machine learning algorithms, including generalized linear models, random forests, extremely randomized trees, gradient-boosted trees, and XGBoost.
• Trained and evaluated models on a substantial cohort of 121,498 pulmonary function tests.

Main Results:

• Machine learning models demonstrated robust performance in estimating lung volumes and capacities.
• Low root mean square error and mean absolute error were observed across predicted lung volumes.
• High area under the receiver operating characteristic curve values (0.81-0.99) indicated strong discriminatory capacity.

Conclusions:

• AI-based spirometry analysis shows significant potential for clinical application.
• These models can aid in the accurate diagnosis and prognosis of respiratory conditions.
• The approach is particularly valuable in settings with limited access to advanced lung volume measurement tools like body plethysmography.