CT Imaging With Machine Learning for Predicting Progression to COPD in Individuals at Risk

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Identifying individuals at risk of Chronic Obstructive Pulmonary Disease (COPD) progression is crucial for timely intervention.
• Early detection can potentially slow disease advancement or facilitate the selection of patient subgroups for novel therapeutic development.

Purpose of the Study:

• To evaluate if combining CT imaging features, radiomic features, and quantitative CT scans with conventional risk factors enhances COPD progression prediction in smokers using machine learning.

Main Methods:

• Utilized data from the Canadian Cohort Obstructive Lung Disease (CanCOLD) study, including CT imaging, spirometry, and demographic data.
• Trained machine learning models to predict COPD progression, comparing models with varying combinations of features.
• Assessed model performance using the area under the receiver operating characteristic curve (AUC).

Main Results:

• Machine learning models incorporating CT imaging features alongside demographics showed improved COPD progression prediction (AUC, 0.730) compared to demographics alone (AUC, 0.649).
• The addition of CT imaging features and spirometry to demographic data further significantly enhanced predictive performance (AUC, 0.877).
• A significant portion of at-risk participants (23.7% in training, 23.0% in testing) progressed to spirometric COPD within 2.5 years.

Conclusions:

• Heterogeneous lung structural changes in at-risk individuals can be quantified using CT imaging features.
• Integrating CT imaging features with conventional risk factors substantially improves the prediction of COPD progression.
• This approach holds promise for earlier identification and management of COPD.