COPD presents a significant diagnostic challenge, particularly during acute exacerbations when standard pulmonary function tests may be difficult or unsafe for patients to perform. In a recent retrospective study, researchers developed a machine learning model capable of diagnosing and grading COPD using chest CT imaging alone. This advancement holds promise for faster, non-invasive assessment of lung function, especially in situations where traditional tests are unfeasible. Notably, the model achieved an accuracy of over 95% in internal testing for diagnosis.
The study involved 173 patients with COPD and 176 healthy controls, recruited between December 2017 and June 2023. Researchers used deep learning segmentation tools to isolate key anatomical features from CT scans, including lung parenchyma, airways, pulmonary arteries, and veins. Statistical feature selection was performed using the Mann-Whitney U-test (p<0.05), followed by the least absolute shrinkage and selection operator (LASSO) to refine the input features. A support vector machine (SVM) classifier was then trained on the internal dataset and evaluated on both internal and external test sets. The external testing cohort included 68 individuals.
The diagnostic model demonstrated excellent performance, with an area under the curve (AUC) of 0.981 in the training set and 0.977 in the internal testing set, alongside corresponding accuracies of 94.9% and 95.6%. For severity grading, the AUC values were 0.889 and 0.796 in the training and testing sets respectively, with accuracies of 78.4% and 71.9%. These findings support the potential of CT-based machine learning models to serve as reliable surrogates for lung function testing in both diagnostic and stratification contexts.
While the results are highly encouraging, the study’s retrospective design and relatively modest external validation cohort limit generalisability. Further large-scale, prospective studies will be essential to validate the model’s clinical utility across diverse populations and imaging protocols. Nonetheless, the research highlights a promising avenue for streamlining COPD diagnosis and management, particularly in acute care settings where conventional testing is impractical. Integration of such models into clinical practice could significantly enhance diagnostic efficiency and decision-making, offering earlier interventions and improved outcomes for patients with COPD.
Reference
Sui H et al. Diagnosis and Severity Assessment of COPD Based on Machine Learning of Chest CT Images. Int J Chron Obstruct Pulmon Dis. 2025;20:2853-67.
