Background: The percentage of gas trapping in computed tomography images acquired at full-expiration (eCT) is associated with COPD symptoms and prognosis, yet it is unknown whether measurements of voxel spatial distribution in eCT have the same associations.

Aim: To investigate the association between texture-radiomics on eCT with established pulmonary function measures of lung hyperinflation and COPD symptoms, and compare these emerging features to existing measures of CT gas trapping.

Methods: Subjects from CanCOLD performed eCT and plethysmography for measurement of residual volume to total lung capacity at baseline (RV/TLC) and 3-year follow-up (?RV/TLC). St. George's Respiratory Questionnaire (SGRQ), Medical Research Council Dyspnea Scale (MRC), and COPD Assessment Test (CAT) assessed COPD symptoms. BODE index assessed COPD prognosis. Low attenuation below -856HU (LAA₈₅₆) quantified percentage of CT gas trapping. Coefficients from a LASSO regression with 102 features of six texture-radiomics sets were summed to create a representative variable (RadScore). Multiple linear regression determined associations of RV/TLC and ?RV/TLC with RadScore and LAA₈₅₆. Binary logistic regression determined associations of symptom scores and COPD prognosis with CT measurements.

Results: 1111 subjects were investigated (n=559 no COPD, n=552 COPD). In models for RV/TLC and ?RV/TLC, RadScore was significant (p<0.01) but LAA₈₅₆ was not. Similarly, in models for SGRQ, MRC, CAT and BODE, only RadScore was statistically significant (p<0.01).

Conclusions: Expiration CT texture-based radiomics have the potential to be an important biomarker when quantifying gas trapping, symptoms, and disease prognosis of COPD.