Assessment of Pulmonary Thromboembolism on Chest Radiographs Using Quantitative AI-Based Cardiovascular-Border Analysis.
Authors
Abstract
To address the utility of an AI-based cardiovascular border analysis of chest radiographs (CXRs) in patients with pulmonary thromboembolism (PTE). This retrospective single-center study included patients with PTE diagnosed using cardiothoracic ratio (CTR) pulmonary angiography between January 2023 and June 2024, along with age-matched and sex-matched controls. The AI model quantified distances from the midline to the cardiovascular borders, including the superior vena cava/ascending aorta (SVC/AO), right atrium (RA), aortic arch, pulmonary trunk (PT), left atrial appendage (LAA), and left ventricle (LV), as well as the CTR. Measurements were standardized as age-adjusted and sex-adjusted Z-scores. Diagnostic performance was assessed using receiver operating characteristic (ROC) analysis. In total, 390 participants (195 with PTE and 195 controls; median age, 68 years) were included in the study. Patients with PTE showed greater values for the PT, SVC/AO, LAA, LV, and CTRs than the controls (<i>p</i> < 0.05). Among 165 patients with follow-up CXRs, post-treatment reductions in values for the PT, SVC/AO, RA, LAA, LV, and CTRs were observed (<i>p</i> < 0.05). ROC analysis showed modest discriminative performance, with the PT Z-score demonstrating the highest area under the curve (AUC, 0.658), improving further in the central PTE subgroup (AUC, 0.700). AI-based cardiovascular-border analysis of CXRs demonstrated modest diagnostic performance but may support treatment monitoring in patient with PTE.