Differentiating Asthma and COPD With CT Imaging and Machine Learning

Can machine learning differentiate asthma and COPD based on computed tomography imaging features, and if so, what subset of CT imaging features should be used?

With machine learning, asthma and chronic obstructive pulmonary disease (COPD) can be differentiated with moderate to high accuracy using 7 computed tomography (CT) features, according to study findings published recently in the European Respiratory Journal.

Researchers sought to determine the optimal subset of CT imaging features for differentiating asthma and COPD with machine learning. To accomplish this, researchers obtained CT imaging from 47 asthma patients and 48 COPD patients recruited from the thorax clinic at Heidelberg University Hospital, Heidelberg, Germany.  Researchers then extracted and analyzed 93 features from the CT images. Differences between the asthma and COPD cohorts for age (P =.25), BMI (P =.31), forced expiratory volume in 1 second (FEV1; P =.31), and for pulmonary function test measurements (P >.05) were not significant. Python and MATLAB were used to implement machine learning algorithms.

Researchers analyzed the data using 2 models: 1 model that included all quantitative CT features, and another model that included only airway features; this was done to clarify the role of CT emphysema and airway remodeling in COPD and asthma differentiation. The investigators found that the analysis considering all quantitative CT features differentiated COPD and asthma using only 7 CT features with high accuracy (accuracy=80%, F1-score=81). In contrast, analysis that considered only CT airway features differentiated COPD and asthma using 8 features with moderate accuracy (accuracy=66%, F1-score=68). Researchers further found that CT emphysema was one of the most important features for distinguishing COPD from asthma.

Researchers concluded that COPD and asthma could be differentiated using machine learning with moderate-high accuracy using 7 CT features: low attenuation area for lower than -950 Hounsfield (LAA950); average out perimeter; average inner perimeter; total airway count; average outer area RB1; average inner area RB1; and total hole count.

Investigators further added, “While CT emphysema was a key factor for distinguishing patients with COPD from asthma, we also showed that CT airway features, such as the total airway count, play an important role in disease differentiation. This can further help to better characterize COPD-asthma-overlap syndromes and identify treatable targets in the future.”

Study limitations included: the small number of participants; expiratory CT images were not collected; the machine learning algorithms may create models that include measurements with no clinical relevance in asthma or COPD; different CT systems with different acquisition parameters were investigated.

Disclosure: One study author declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.


Moslemi A, Kontogianni K, Brock J, Wood S, Herth F, Kirby M. Differentiating COPD and asthma using quantitative CT imaging and machine learning. Eur Respir J. Published online February 24, 2022. doi:10.1183/13993003.03078-2021

This article originally appeared on Pulmonology Advisor