Background/Purpose: Radiomics describes the in-depth analysis of tissue phenotypes by computational retrieval of high-dimensional quantitative imaging features including tissue intensity, texture, and wavelet characteristics. Here, we aimed to evaluate high-resolution computed tomography (HRCT)-based radiomics for disease phenotyping and risk stratification in interstitial lung disease related to systemic sclerosis (SSc-ILD).

Methods: In this study, we investigated two independent, prospectively followed SSc-ILD cohorts including 90 patients (76.7% female, median age 57.5 years) from the University Hospital Zurich and 66 patients (75.8% female, median age 61.0 years) from the Oslo University Hospital. For every subject, we defined and extracted 1,386 radiomic features from semi-automated segmented HRCT images including 17 intensity, 137 texture, and 1,232 wavelet features using our in-house developed radiomics software Z-Rad. After filtering of robust radiomic features, we performed unsupervised k-Means clustering and supervised prediction modelling to 1) identify homogeneous imaging-based groups without any a priori assumptions, and 2) to derive a quantitative radiomic risk score for progression-free survival (PFS) in SSc-ILD. PFS was defined as the time to a relative decline in FVC% predicted ≥ 15%. Associations with clinical characteristics at baseline and PFS among the obtained clusters and risk score groups were assessed by Fisher’s Exact and Mann-Whitney U test, or univariable Cox regression, respectively.

Results: Unsupervised cluster analysis of 1,355 robust radiomic features revealed two distinct patient clusters based on their radiomic profiles. The two patient clusters exhibited significant differences in their lung disease-specific baseline parameters, but not in serological or demographic characteristics. Cluster 2 presented a more severe ILD phenotype than cluster 1, and was significantly associated (p< 0.05) with a worse restrictive ventilation defect, pulmonary hypertension, a fibrosis extent on HRCT of ≥ 20%, and certain visual HRCT ILD patterns, including UIP radiological subtype and honeycombing. The clusters further significantly differed in their outcome with cluster 2 showing a decreased PFS and thus a higher risk of faster disease progression (p=0.001). We next derived a quantitative radiomic risk score (qRISSc) composed of the sum of 26 equally weighted radiomic features that accurately predicted PFS and showed prognostic power in both study cohorts (C-index = 0.67 for Zurich, 0.71 for Oslo). We also compared the prognostic potential of qRISSc to existing SSc-ILD stratification tools, including subgrouping of patients based on HRCT (< 20% or ≥20% fibrosis) or the FVC% predicted threshold of < 70%, respectively. In both cohorts, neither HRCT- nor FVC%-based risk stratification was prognostic for future lung function decline, overall indicating the superiority of qRISSc over current prognostic measures.

Conclusion: Our data suggests that radiomic features and radiomics-derived scores can capture important phenotypic and prognostic information thus showing great potential for risk stratification in SSc-ILD.

« Back to ACR Convergence 2020

ACR Meeting Abstracts - https://acrabstracts.org/abstract/resolving-phenotypic-and-prognostic-differences-in-interstitial-lung-disease-related-to-systemic-sclerosis-by-computed-tomography-based-radiomics/