Author: Jean Bourbeau, Jim Hogg, Miranda Kirby, Meghan Koo, Kalysta Makimoto, Wan Tan 👨🔬
Affiliation: Montreal Chest Institute of the Royal Victoria Hospital, McGill University Health Centre, Toronto Metropolitan University, Centre for Heart Lung Innovation, University of British Columbia 🌍
Purpose: Chronic obstructive pulmonary disease (COPD) exacerbations are burdensome to patients and healthcare systems. CT imaging-derived measures of emphysema and airway remodeling have been shown to predict COPD exacerbations. The Jacobian determinant is a whole-lung average measurement of biomechanical contraction and expansion, however it is not sensitive to focal abnormalities. Previous work has shown that Jacobian-derived texture-based radiomics feature large-distance-high-gray-level-emphasis (LDHGLE) improved model performance compared to the first-order Jacobian measurement for lung function and lung function decline. Our objective was to evaluate the predictive performance of Jacobian-LDHGLE for COPD exacerbations at follow-up.
Methods: CanCOLD participants performed CT imaging at full-inspiration/expiration at baseline and reported exacerbation frequency and severity at 3-year follow-up. CT emphysema and airway remodeling measures were extracted from full-inspiration images; the Jacobian was extracted using registered inspiration-expiration images; Jacobian values were discretized for second-order texture-based radiomics extraction. Negative binomial regression models with clinical and imaging covariates were constructed to evaluate the predictive performance of first-order Jacobian and second-order Jacobian-LDHGLE for symptom-based, event-based, and moderate-severe exacerbations at follow-up.
Results: 719 participants were investigated (n=271 at-risk, n=267 mild-COPD, n=181 moderate-severe COPD). In independent models, the first-order Jacobian was only able to predict future symptom-based exacerbations (p<0.05), however, the second-order Jacobian-LDHGLE was able to predict future symptom-based, event-based and moderate-severe exacerbations (p<0.05). When placed in the same model, the second-order Jacobian-LDHGLE remained statistically significant in predicting future exacerbations (p<0.05). When placed in the same model as CT emphysema and airway remodeling, the second-order Jacobian-LDHGLE remained statistically significant in predicting future symptom-based and moderate-severe exacerbations (p<0.05).
Conclusion: This work provides insights on the role of heterogenous abnormal lung biomechanics in COPD exacerbations. CT lung measurements such as the second-order Jacobian-LDHGLE has the potential to be an important biomarker in personalized and targeted COPD prognosis care.