Background: Pulmonary hypertension (PH) has been reported in 8-15% of patients with idiopathic pulmonary fibrosis (IPF) upon initial work-up and is an important predictor of mortality. Presenting symptoms are non-specific, leading to a delayed clinical suspicion and diagnosis. Widely available non-invasive tools for upfront screening are therefore needed, both to detect PH in IPF and to predict severe PH IPF that need referral for further work-up, including right heart catheterization (RHC), at specialist PH centres.

Aims and Objectives: To build a framework of deep learning (DL) models to diagnose and predict severity of PH in patients with IPF.

Methods: This is a pilot, retrospective, multi-centre, case-control study based on routine clinical data between 2005 and 2022. Below we present our study protocol.

Results: Approximately 75 adult patients with PH-IPF and 75 controls (IPF without PH) matched for age, sex and forced vital capacity will be included. PH will be confirmed with RHC according to latest guidelines and excluded with echocardiography and/or RHC. Echocardiography and computed tomography (CT) raw data, spirometry parameters and brain natriuretic peptide (BNP) will be used (Figure). The accuracy of the DL classification algorithm will be evaluated.

Conclusions: We hope to propose utilization of DL algorithms to detect PH in patients with IPF based on raw non-invasive data acquired at expert centres to enable early PH detection and PH severity classification.