Abstract Rationale Hypersensitivity pneumonitis (HP) is an immune-mediated interstitial lung disease that occurs in individuals who are both susceptible and sensitized to inhaled antigens, and various disease behaviors and prognosis. Identifying imaging-derived phenotypes that reflect disease activity and prognosis remains an unmet need. We hypothesized that quantitative radiomics analysis of high-resolution CT (HRCT) could stratify patients into distinct phenotypes associated with clinical and prognostic differences. Methods We retrospectively analyzed 182 consecutive patients with multidisciplinary discussion-diagnosed fibrotic HP. Quantitative features were extracted from HRCT using PyRadiomics and standardized by z-scoring. Unsupervised clustering was performed using K-means (k = 2). Clusters were compared regarding quantitative CT obtained by Q-ZIP (Ziosoft Informatics Platform), clinical information, physiologic parameters (FVC, DLco) and serum biomarkers. Overall survival (OS) was analyzed by Kaplan-Meier and Cox proportional hazards models. Results Radiomics analysis stratified the cohort into two reproducible phenotypes: Cluster 0 (n = 108) and Cluster 1 (n = 74). Cluster 1 exhibited higher image heterogeneity (entropy, short-run emphasis, run-length non-uniformity) and lower uniformity, consistent with a complex patchy texture pattern. Q-ZIP confirmed normal lung ratio with higher volumes of ground-glass opacity, reticulation, and traction bronchiectasis (p 0.001 for all) in Cluster 1. Clinically, Cluster 1 showed greater physiologic impairment with %FVC 74.6 ± 17.1 vs 96.8 ± 17.1 (p 0.001), DLco 11.1 ± 3.3 vs 15.7 ± 4.4 mL/min/mmHg (p 0.001), and higher serum KL-6 (1984 ± 1944 vs 958 ± 721 U/mL, p 0.001). BMI was modestly higher (25.6 vs 23.9, p = 0.009) in Cluster 1. Prognosis: Cluster 1 had significantly worse OS (log-rank p = 0.004). Unadjusted Cox analysis demonstrated increased mortality (HR 2.55, 95% CI 1.31-4.95). After adjustment for age, FVC, DLco, and KL-6, the association attenuated (HR 1.68, p = 0.25), indicating overlap with physiologic severity. The cluster remained stable after feature exclusion (adjusted Rand Index 0.89; 97% agreement). Conclusion Quantitative HRCT radiomics objectively stratified fibrotic HP into two imaging phenotypes with distinct physiologic and prognostic characteristics. The high heterogeneity phenotype reflected increased fibrotic burden, reduced DLco, elevated serum KL-6, and poorer survival, corresponding to a progressive fibrotic phenotype. Although cluster effect diminished after physiologic adjustment, radiomics provides a reproducible, non-invasive marker that complements pulmonary function and quantitative CT for individualized management of fibrotic HP. This abstract is funded by: None
Matsuura et al. (Fri,) studied this question.