Light Convolutional Neural Network to Detect Chronic Obstructive Pulmonary Disease (COPDxNet): A Multicenter Model...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) affects a large proportion of adults, with many remaining undiagnosed.
• Opportunistic screening via chest computed tomography (CT) scans presents a viable strategy for improving COPD detection rates.
• Deep-learning models offer a pathway to develop simple, clinically applicable tools for COPD identification.

Purpose of the Study:

• To develop and validate a deep-learning model for detecting COPD using minimally processed chest CT scans.
• To assess the model's performance on both standard-dose and low-dose CT scans.
• To evaluate the generalizability of the model across different patient cohorts and imaging protocols.

Main Methods:

• A lightweight convolutional neural network, COPDxNet, was developed.
• The model was trained on 13,043 chest CT scans from the COPDGene study, randomly split into training and testing sets.
• External validation was performed on datasets from SPIROMICS and the National Lung Screening Trial (NLST).

Main Results:

• COPDxNet achieved high accuracy in detecting COPD on standard-dose CT scans (AUC 0.92) and low-dose scans (AUC 0.88) within the COPDGene cohort.
• The model demonstrated strong performance on external validation datasets, with AUCs of 0.92 for SPIROMICS and 0.82 for NLST.
• The model was well-calibrated across all datasets, indicating reliable probability estimates.

Conclusions:

• COPDxNet exhibits high discriminative accuracy and generalizability for COPD detection on various chest CT scan types.
• The model's performance supports its potential utility in clinical practice and screening programs for diverse populations.
• This deep-learning approach can aid in identifying undiagnosed cases of COPD, improving patient outcomes.