Accuracy of Deep Learning in Diagnosing Chronic Obstructive Pulmonary Disease: Systematic Review and Meta-Analysis

Area of Science:

• Artificial Intelligence in Medicine
• Pulmonary Medicine
• Machine Learning Applications

Background:

• Chronic obstructive pulmonary disease (COPD) is a prevalent respiratory condition.
• Deep learning (DL) shows promise for COPD diagnosis and severity grading.
• Limited systematic evidence exists on the accuracy of DL for COPD detection and grading.

Purpose of the Study:

• To systematically evaluate the diagnostic and grading accuracy of DL models for COPD.
• To provide evidence for developing intelligent COPD diagnostic tools.
• To synthesize current evidence on DL applications in COPD management.

Main Methods:

• Systematic literature search across major databases (PubMed, Embase, etc.) up to November 2025.
• Meta-analysis of 56 studies (886,753 participants) using bivariate and random-effects models.
• Risk of bias assessment using QUADAS-2; subgroup analyses by data source and validation methods.

Main Results:

• DL models achieved high accuracy for binary COPD detection (AUC=0.93), particularly with CT (AUC=0.92) and breath sounds (AUC=0.98).
• Pooled sensitivity and specificity for COPD detection were 0.87 and 0.88, respectively.
• DL models demonstrated limited accuracy in discriminating between Global Initiative for Chronic Obstructive Lung Disease (GOLD) stages.

Conclusions:

• DL models show strong potential for COPD diagnosis using CT and audio data.
• Current DL models struggle with accurate multiclass grading of COPD severity (GOLD stages).
• Further high-quality, multicenter studies are needed to address heterogeneity and improve external validation.