Background: The analysis of cardiovascular borders (CVBs) in chest x-rays (CXRs) traditionally relied on subjective assessment and does not have established normal ranges.
Objectives: The authors aimed to develop a deep learning-based method for quantifying CVBs on CXRs and to explore its clinical utility.
Methods: This study used a prevalidated deep learning to analyze CVBs. A total of 96,129 normal CXRs from 4 sites were used to establish age- and sex-specific normal ranges of CVBs. The quantified CVBs were standardized into z-scores for newly inputted CXRs. The clinical utility of the z-score analysis was tested using 44,567 diseased CXRs from 3 sites (9,964 valve disease; 32,900 coronary artery disease; 1,299 congenital heart disease; 294 aortic aneurysm; 110 mediastinal mass).
Results: For distinguishing valve disease from normal controls, the area under the receiver operating characteristic curve for the cardiothoracic ratio was 0.80 (95% CI: 0.80-0.80), while the combination of right atrium and left ventricle borders had an area under the receiver operating characteristic curve of 0.83 (95% CI: 0.83-0.83). Between mitral and aortic stenosis, z-scores of CVBs were significantly different in the left atrial appendage (1.54 vs 0.33, P < 0.001), carinal angle (1.10 vs 0.67, P < 0.001), and ascending aorta (0.63 vs 1.02, P < 0.001), reflecting disease pathophysiology. Cardiothoracic ratio was independently associated with a 5-year risk of death or myocardial infarction in the coronary artery disease (z-score ≥2, adjusted HR: 3.73 [95% CI: 2.09-6.64], reference z-score <-1).
Conclusions: Deep learning-derived z-score analysis of CXR showed potential in classifying and stratifying the risk of cardiovascular abnormalities.
Keywords: artificial intelligence; cardiovascular borders; cardiovascular disease detection; chest x-rays.
Copyright © 2025 The Authors. Published by Elsevier Inc. All rights reserved.