The Brainomix automated e-ILD CT algorithm outperforms forced vital capacity in predicting outcomes for patients with idiopathic pulmonary fibrosis

Objectives: e-ILD (Brainomix (Oxford, UK)) is an AI-powered image processing module which quantifies thoracic CT biomarkers of patients with interstitial lung disease. Using the Open-Source Imaging Consortium database, associations between e-ILD biomarkers and clinical outcomes were explored. Methods: Analyses were restricted to IPF patients with contemporaneous CT scans and lung function tests. The input to e-ILD is the 3D image data from a single CT; derived outputs include a novel imaging biomarker, the weighted reticulovascular score (WRVS). Relationships between imaging biomarkers, lung function and survival were analysed using Cox proportional hazards regression and Harrell's C-index. Results: Data from 278 IPF patients were analysed, 104 with sequential imaging (mean interval 54 weeks). Adjusting for lung function, C-index for transplant free survival was 0.75 for baseline WRVS and 0.66 for forced vital capacity (FVC). Patients were divided into low and high WRVS groups at the median. Mortality was four times greater in the high WRVS group (HR 4.0 CI 2.8-5.7, p<0.001). FVC was half as prognostic when dichotomised at the median FVC of 80% (HR 2.0 CI 1.4-2.8, p<0.001). Survival prediction improved when incorporating change in WRVS (C-index 0.63) to change in FVC (0.56). Baseline WRVS predicted future FVC decline≥10% (OR 5.8 CI 1.5–22.3, p=0.01). Discussion: From a baseline CT scan, e-ILD automated biomarkers predict FVC decline and survival in IPF patients outperforming change in FVC. Exploratory clinical trial endpoints combining automated imaging biomarkers with FVC may improve patient selection and definition of treatment response.