Developing an AI-Powered System for Enhanced Pulmonary Hypertension Imaging Assessments

Pulmonary hypertension (PH) is a severe condition that is commonly diagnosed late and is associated with high mortality. Imaging plays a key role in the work up of patients with suspected PH as described in international guidelines. Currently imaging approaches in PH rely on visual assessments or manual measurements. Our pilot data demonstrates that artificial intelligence (AI) assessments have greater clinical value. There is an urgent need to develop a system to develop explainable AI, optimise implementation, to improve patient assessments. The three key aims of this fellowship are to use AI approaches to identify new diagnose PH earlier, identify new therapy responsive phenotypes and improve prognostication. The AI system involves training of highly heterogeneous imaging data, with contours and labels made by experts. Unsupervised and supervised AI approaches shall be employed to automate explainable pathophysiological and haemodynamic measurements, and more deeply phenotype patients from uni and multimodal data with thorough validation including scan-rescan testing. The proposed AI system include explainable models with human interaction to improve explainability, generalisability and accuracy, and can improve the detection and management of patients with PH, and provides an approach that translates to other heart and lung conditions.