Enhancing COPD Classification Using Combined Quantitative CT and Texture-Based Radiomics: A CanCOLD Cohort Study

Abstract

Recent advancements in texture-based CT radiomics have demonstrated its potential for classifying chronic obstructive pulmonary disease (COPD).Participants from Canadian Cohort Obstructive Lung Disease (CanCOLD) were investigated. A total of 108 features were included: 8 qCT, 95 texture-based radiomics, and 5 demographics. Machine learning models included demographics along with texture-based radiomics, and/or qCT. Combinations of 5 feature-selection and 5 classification methods were evaluated; a training dataset was used for feature-selection and to train the models, and a testing dataset was used for model evaluation. Models for classifying COPD status and severity were evaluated using the area under the receiver operating characteristic curve (AUC) with DeLong's test for comparison. SHAP analysis was used to investigate the features selected.A total of 1204 participants were evaluated (n=602 no COPD; n=602 COPD). There were no differences between the groups for sex (p=.77) or BMI (p=.21). For classifying COPD status, the combination of demographics, texture-based radiomics and qCT obtained higher performance (AUC=0.87) compared to demographics and texture-based radiomics (AUC=0.81; p<.05) or qCT (AUC=0.84; p<.05). Similarly, for classifying COPD severity, the combination of demographics, texture-based radiomics and qCT obtained higher performance (AUC=0.81) compared to demographics and texture-based radiomics (AUC=0.72; p<.05) or qCT (AUC=0.79; p=<.05). Texture-based radiomics and qCT features were among the top 5 features selected (15th-percentile-of-the-CT-density-histogram, CT total-airway-count, pack-years, CT grey-level-co-occurrence-matrix zone-distance-entropy, CT low-attenuation-clusters) for classifying COPD status.Texture-based radiomics and conventional qCT features in combination improve machine learning models for COPD classification of status and severity.