Artificial Intelligence and Chest Computational Tomography to predict prognosis in Pulmonary Hypertension and lung disease.

Pulmonary hypertension (PH) is an incurable severe condition with poor survival and multiple clinically distinct sub-groups and phenotypes. Accurate diagnosis and identification of the underlying phenotype is an integral step in patient management as it informs treatment choice. Outcomes vary significantly between phenotypes. Patients presenting with signs of both PH and lung disease pose a clinical dilemma between two phenotypes - idiopathic pulmonary arterial hypertension (IPAH) and pulmonary hypertension secondary to lung disease (PH-CLD) as they can present with overlapping features. The impact of lung disease on outcomes is not well understood and this is a challenging area in the literature with limited progress. All patients suspected with PH undergo routine chest Computed Tomography Pulmonary Angiography (CTPA) imaging. Despite this, the prognostic significance of commonly visualised lung parenchymal patterns is currently unknown. Current radiological assessment is also limited by its visual and subjective nature. Recent breakthroughs in deep-learning Artificial Intelligence (AI) approaches have enabled automated quantitative analysis of medical imaging features. This thesis demonstrates the prognostic impact of common lung parenchymal patterns on CT in IPAH and PH-CLD. It describes how this data could aid in phenotyping, and in identification of new sub-groups of patients with distinct clinical characteristics, imaging features and prognostic profiles. It further develops and clinically evaluates an automated CT AI model which quantifies the percentage of lung involvement of prognostic lung parenchymal patterns. Combining this AI model with radiological assessment improves the prognostic predictive strength of lung disease severity in these patients.