Enabling chronic obstructive pulmonary disease diagnosis through chest X-rays: A multi-site and multi-modality study.

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Medical Imaging

Background:

• Chronic Obstructive Pulmonary Disease (COPD) is a prevalent global respiratory illness.
• Early diagnosis of COPD is challenging, often leading to delayed or ineffective treatment.
• Spirometry, the current diagnostic standard, faces accessibility issues, particularly in resource-poor settings.

Purpose of the Study:

• To develop and validate deep learning (DL) models for COPD detection using chest X-rays (CXRs).
• To explore the potential of CXRs as a screening tool for identifying patients who may have COPD.
• To assess model performance across different demographic groups for fairness.

Main Methods:

• Utilized three CXR datasets and associated electronic health records (EHR).
• Developed two DL fusion schemes: model-level (Bootstrap aggregating) and data-level (multi-modal CXR and EHR data).
• Conducted fairness analysis to evaluate model performance across demographic subgroups.

Main Results:

• Deep learning models achieved an Area Under the Curve (AUC) greater than 0.75 in detecting COPD from CXRs.
• The developed models show promise for facilitating COPD patient screening.
• CXRs present a viable alternative for COPD screening in regions with limited access to spirometry.

Conclusions:

• Chest X-rays, a widely available diagnostic test, can be leveraged for early COPD detection.
• Further research can build upon these findings to improve early diagnosis and treatment of COPD.
• This approach has the potential to alter the disease course for many patients, particularly those undiagnosed or undertreated.