AutoCOPD-A novel and practical machine learning model for COPD detection using whole-lung inspiratory quantitative CT...

Area of Science:

• Pulmonary Medicine
• Radiology
• Medical Imaging Analysis

Background:

• Chronic obstructive pulmonary disease (COPD) diagnosis rates are globally low.
• Quantitative computed tomography (QCT) parameters offer valuable insights into airway and lung parenchyma alterations in COPD.
• This study focuses on a whole-lung inspiratory CT model for COPD detection.

Purpose of the Study:

• To assess the performance of QCT features in detecting COPD.
• To develop and validate a novel multimodal framework for COPD detection using QCT.
• To evaluate the generalizability and feasibility of the developed model across diverse clinical settings.

Main Methods:

• A multicenter retrospective study involving 4106 participants.
• Development of a multimodal framework using eXtreme gradient boosting and hybrid feature selection.
• Validation using derivation and three external cohorts, including the National Lung Screening Trial (NLST) for low-dose CT scans.

Main Results:

• The AutoCOPD model, with ten QCT features, achieved an AUC of 0.860 in internal validation.
• External validation demonstrated excellent discrimination across multiple cohorts (AUCs ranging from 0.881 to 0.915).
• Decision curve analysis confirmed the clinical utility of AutoCOPD across various COPD risk thresholds.

Conclusions:

• The AutoCOPD model effectively identifies heterogeneous COPD using a limited set of QCT features.
• This model shows potential for early detection of mild or asymptomatic COPD, addressing delayed diagnoses.
• AutoCOPD is a feasible tool for routine practice, generalizable across clinical settings.