Automated, Standardized, Quantitative Analysis of Cardiovascular Borders on Chest X-Rays Using Deep Learning

doi:10.1016/j.jacadv.2025.101687

Automated, Standardized, Quantitative Analysis of Cardiovascular Borders on Chest X-Rays Using Deep Learning

JACC AdvPublished 25 Apr 2025development of a deep learning-based method for quantifying cardiovascular borders (CVBs) on chest x-rays (CXRs)

Read the original article ↗

Cite

Authors' conclusion

Deep learning-derived z-score analysis of CXR showed potential in classifying and stratifying the risk of cardiovascular abnormalities.

Study context

Population: Normal and diseased chest x-rays (valve disease, coronary artery disease, congenital heart disease, aortic aneurysm, mediastinal mass)CXRs, 96,129 normal and 44,567 diseased · normal CXRs (4 sites) 96,129 · diseased CXRs (3 sites) 44,567 · valve disease 9,964 · coronary artery disease 32,900 · congenital heart disease 1,299 · aortic aneurysm 294 · mediastinal mass 110

Comparison: z-score analysis of CVBs in diseased CXRs against normal ranges
normal controlsn = 96,129
valve diseasen = 9,964
coronary artery diseasen = 32,900
congenital heart diseasen = 1,299
aortic aneurysmn = 294
mediastinal massn = 110

Methodological detailsprevalidated deep learning, 4 sites (normal CXRs), 3 sites (diseased CXRs), age- and sex-specific normal ranges, z-scores

Prevalidated deep learning
4 sites (normal CXRs)
3 sites (diseased CXRs)
Age- and sex-specific normal ranges
Z-scores

Main finding

As reported in the abstract

Distinguishing valve disease from normal controls

AUROC, cardiothoracic ratio

0.80

95% CI 0.80–0.80

AUROC, combination of right atrium and left ventricle borders

0.83

95% CI 0.83–0.83

Further findings

Outcome	Values	← Lower risk with z-score ≥2 ←→ Higher risk with z-score ≥2 →	Adjusted HR (95% CI)
Death or myocardial infarction by cardiothoracic ratio, coronary artery disease · 5-year · independently associated		0.10.31310	3.73 (2.09–6.64)
		0.10.31310Adjusted hazard ratio (log scale)

Outcome	Values	Estimate
Z-scores of CVBs, mitral vs aortic stenosis	mitral stenosis left atrial appendage 1.54 · carinal angle 1.10 · ascending aorta 0.63 · aortic stenosis left atrial appendage 0.33 · carinal angle 0.67 · ascending aorta 1.02

Z-scores of CVBs, mitral vs aortic stenosis: left atrial appendage P < 0.001
Z-scores of CVBs, mitral vs aortic stenosis: carinal angle P < 0.001
Z-scores of CVBs, mitral vs aortic stenosis: ascending aorta P < 0.001

Original publication

Automated, Standardized, Quantitative Analysis of Cardiovascular Borders on Chest X-Rays Using Deep Learning. JACC Adv. Published 25 April 2025. doi:10.1016/j.jacadv.2025.101687

Read the original article ↗PubMed

Summarised from the published abstract; every number is checked against it