e-Lung (Brainomix, Oxford) is an AI-powered image processing module which quantifies thoracic CT biomarkers of patients with interstitial lung disease (ILD) and may be more prognostic than forced vital capacity (FVC). Using data from the Open-Source Imaging Consortium, we sought to validate previously reported associations between e-Lung imaging biomarkers and clinical outcomes.

3D image data from CT scans in non-IPF fibrotic ILD patients with contemporaneous lung function tests were used. Derived outputs include a novel imaging biomarker, the weighted reticulovascular score (WRVS), which incorporates fibrotic features and abnormal vascular structures. Analyses were performed using Cox proportional hazards regression and Harrell?s C-index with 95% confidence intervals.

Data from 352 patients were analysed (Cohort 1 n=182, Cohort 2 n=170). Adjusting for age and lung function, in individual cohorts and when pooled, the WRVS from a baseline HRCT scan was better associated with transplant free survival than FVC (pooled c-indices 0.72 v 0.65). When dichotomised at the medians, a high WRVS was more prognostic (HR 3.6 CI 2.4-5.3 p<0.001) than low FVC (HR 2.4 CI 1.6-3.5 p<0.001). Baseline WRVS was more than twice as predictive of future FVC decline?10% over 12 months (OR 5.3 CI 2.6-11.8 p<0.001) than baseline FVC (OR 2.2 CI 1.2-4.4, p=0.019).

In two independent cohorts, e-Lung automated biomarkers are highly prognostic, outperforming FVC confirming previous findings in patients with IPF. Clinical trial endpoints combining e-Lung imaging biomarkers with FVC may improve definition of treatment response in ILD; this is being explored in prospective clinical trials.