Radiomic Profiling of Chest CT in a Cohort of Sarcoidosis Cases

Background High resolution computed tomography (HRCT) of the chest is increasingly used in clinical practice for sarcoidosis. Visual assessment of chest HRCTs in patients with sarcoidosis has high inter- and intra-rater variation. Radiomics offers a reproducible quantitative assessment of HRCT lung parenchyma and could be useful as an additional summary measure of disease. We develop radiomic profiles on HRCT and map them to radiologic, clinical, and patient reported outcomes. Research Question Can radiomic analysis of chest HRCT cluster patients into groups that are related to radiologic, clinical, and patient reported outcomes? Study Design and Methods Three-dimensional radiomic features were calculated on chest HRCT for both lungs from sarcoidosis cases enrolled in the Genomic Research in Alpha-1 Antitrypsin Deficiency and Sarcoidosis (GRADS) study (N=320). Robust and sparse K-means was used to cluster sarcoidosis cases using their radiomic profiles. Differences in patterns on visual assessment (VAS) by cluster were identified using chi-squared tests. Linear regression investigated how pulmonary function tests and patient reported outcomes differed between clusters with and without adjustment for other radiologic quantification. Results Radiomic-based clustering identified four clusters associated with both Scadding stage and Oberstein score ( P <0.001). One of the clusters had markedly few abnormalities. Another cluster had consistently more abnormalities along with more Scadding stage IV. Average pulmonary function testing (PFT) differed between clusters, even after accounting for Scadding stage and Oberstein score ( P <0.001), with one cluster having more obstructive disease. The most discriminative radiomic measures explained 10-15% of the variation in PFT beyond demographic variables. Shortness of breath, fatigue, and physical health differed by cluster ( P <0.014). Interpretation Radiomic quantification of sarcoidosis identifies new subtypes representative of existing radiologic assessment and more predictive of pulmonary function. These findings provide evidence that radiomics may be useful for identifying new imaging-based disease phenotypes.