Background/Purpose: Interstitial lung disease (ILD) is a major cause of mortality in patients with autoimmune-related CTD such as SSc and RA. Molecular characterization of the lung could benefit disease management but biopsies are highly invasive. High-dimensional biomedical image analysis, known as radiomics, provides quantitative insights into organ-scale pathophysiology by digital disease fingerprinting. We previously reported that radiomics conveys lung functional and molecular phenotypes with important implications for personalized diagnosis and prognostication in SSc-ILD (Schniering et al. Eur Respir J, 2022, 59 (5)). Here, we aimed to integrate time-resolved radiomic and proteomic tissue profiles to identify non-invasive radiomics-based molecular surrogate signatures that allow longitudinal assessment of ILD development.

Methods: C57BL/6J mice (8 weeks old, female) received 2 U/kg bleomycin (n=54) or saline (n=39) on day 0 to model lung fibrosis or healthy conditions, respectively. MicroCT-derived radiomic features (n=1’130) were extracted from semi-automatically segmented lungs  on days 3, 7, 14, 21, 28, 35, and whole lung proteomics was measured. Unsupervised clustering of the Fréchet distances, a measure of similarity between curves, was used to identify radiomic feature trajectories with distinct patterns during disease development. Integration of the resulting clusters with corresponding protein expression profiles by linear regression was performed for functional characterization and derivation of time-resolved radioproteomic association modules (p< 0.05, t statistic >2). Reactome pathway enrichment analysis of the modules was performed to establish the biological basis of the imaging profiles. The identified signatures were validated in independent experimental cohorts. 

Results: Unsupervised clustering of the radiomic feature trajectories revealed 6 clusters, consisting of 94, 194, 202, 325, 137, and 144 features, that mirrored distinct stages of ILD development. Integration with sample-matched proteomics identified 52, 356, 98, 52, 27, and 382 proteins that paralleled the radiomic trajectories in clusters 1 to 6, respectively. Enrichment analysis of the resulting radioproteomic association modules demonstrated that they conveyed distinct disease-relevant pathway activity. These included, amongst others, pro-fibrotic tissue remodeling (e.g. LOXL1, COL5A1, PLO3), hemostasis (e.g. CD36, CD47, ITGB1), and fatty acid metabolism (e.g. ACLS1, ACOT1, HDAH). Evaluation of the radiomic signatures in an independent cohort (days 3, 14, 42) established their reproducibility. Most intriguing, their application in a chronic lung fibrosis model induced by repetitive bleomycin injury allowed non-invasive molecular profiling on imaging level, revealing differential pathway regulation between resolving (single injury) and non-resolving (repetitive injury) lung fibrosis.

Conclusion: Radiomics serves as a non-invasive surrogate to characterize and explore the molecular landscape of the lung during experimental ILD development.

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/radiomics-non-invasively-conveys-time-resolved-molecular-pathway-activity-in-experimental-fibrosing-interstitial-lung-disease/