Multi-modal deep learning and domain knowledge integration to aid multidisciplinary teams in diagnosing in idiopathic pulmonary fibrosis

Idiopathic pulmonary fibrosis (IPF) is a devastating condition where the normal lung parenchyma is replaced by scar tissue. It leads to respiratory failure and ultimately death on average three years from the time of diagnosis, a survival rate worse than many cancers, and it accounts for 1 in every 100 deaths in the UK. There are no current methods of identifying patients likely to have a more aggressive disease phenotype, where early treatment may have greatest benefit: 1/3rd of patients receive>1 misdiagnosis and 50% wait 6 months for a diagnosis leaving patients in limbo with no clear therapeutic strategy. Modelling multi-modal data using AI could reduce variability in diagnoses compared with current multi-disciplinary team assessments.