Lung Density: A Computed Tomography Marker for Severity in Pulmonary Hypertension of Various Etiologies.
TL;DR
Automated non-contrast CT-AI densitometric metrics consistently discriminate severe PVR (>5 Wood units), led by MLD (AUC = 0.686; threshold -785HU), serving as potential exploratory screening indicators that reflect underlying cardiopulmonary alterations.
Abstract
INTRODUCTION To investigate the capabilities of CT markers for assessing the severity of PH and to compare them with PFT and gas exchange parameters.
Methods
This single-center study evaluated 90 consecutive Pulmonary Hypertension (PH) patients across five diagnostic subgroups (idiopathic, connective tissue disease, congenital heart disease, left heart disease, and chronic thromboembolic). Baseline diagnostics included comprehensive clinical, functional, and invasive right heart catheterization. Automated non-contrast chest CT densitometry via the Siemens Syngo.via (VB80B) workstation extracted volumetric parenchymal metrics: Mean Lung Density (MLD), Percentile 75, and High Attenuation Volume (HAV). Statistical robustness was verified using 1000-resample percentile bootstrap simulations, followed by multi-parameter ROC analysis to determine the diagnostic accuracy of CT-AI metrics in identifying severe pulmonary vascular resistance (PVR > 5 Wood units).
Results
In bootstrap-validated multivariable regression models, right atrium area emerged as an independent predictor for both MLD (β = 2.02, 95% CI: 0.21-3.83, p = 0.030) and Percentile 75 (β = 2.19, 95% CI: 0.21-4.17, p = 0.031), while forced vital capacity independently predicted High Attenuation Volume (HAV; β = -0.031, 95% CI: -0.053- -0.009, p = 0.007). ROC curve analysis demonstrated modest but consistent diagnostic performance of non-contrast CT-AI metrics in discriminating severe PVR (>5 Wood units), led by MLD (AUC = 0.686, 95% CI: 0.511-0.860; candidate threshold > -785 HU) and closely mirrored by Percentile 75 (AUC = 0.680) and HAV (AUC = 0.680).
Discussion
The established associations between AI-driven lung density, right atrial remodeling, and spirometric volumes reflect descriptive cross-sectional associations that may parallel underlying macro-structural changes. Furthermore, automated density profiling provides an objective, observer-independent tool to capture mosaic parenchymal alterations, serving strictly as an exploratory adjunct for preliminary phenotype characterization.
Conclusion
Automated non-contrast CT-AI densitometric metrics consistently discriminate severe PVR (>5 Wood units), led by MLD (AUC = 0.686; threshold -785HU), serving as potential exploratory screening indicators that reflect underlying cardiopulmonary alterations. They do not replace established clinical tools but offer potential value for opportunistic screening, warranting further prospective multi-center and prognostic validation.