Quantitative Comparison of Virtual and True Noncontrast Lung Images in Photon-counting CT: Global and Regional Attenuation Assessment.

Abstract

Background Use of virtual noncontrast (VNC) images from photon-counting CT (PCCT) may reduce radiation dose and support quantitative lung analysis by obviating the need for separate true noncontrast (TNC) acquisitions. Their accuracy for lung densitometry, however, has not been systematically assessed. Purpose To evaluate agreement between VNC and TNC images for global and regional lung attenuation measurements in thoracic PCCT. Materials and Methods This retrospective study included patients who underwent same-day unenhanced and contrast-enhanced thoracic PCCT between November 2021 and January 2024. VNC datasets were reconstructed from contrast-enhanced scans. Lung volume, mass, and attenuation were compared using paired t tests, Wilcoxon signed-rank tests, and Bland-Altman analysis. Inspiratory depth effects were assessed with linear regression. Density class distributions were compared, and voxel-wise attenuation differences were evaluated in a subset of patients after nonlinear registration. Results This study included 83 patients (mean age, 60.0 years ± 15.0 [SD]; 47 men). Mean lung volume (4143 mL ± 1375 vs 4111 mL ± 1384; P = .41) and mean attenuation (-799 HU ± 56 vs -798 HU ± 56; P = .91) were similar between VNC and TNC images. After adjustment for inspiratory depth, VNC and TNC images demonstrated no systematic bias, and attenuation differences demonstrated substantially lower variability after adjustment than before adjustment (mean difference, -1.1 HU ± 5.9 vs -0.8 HU ± 19.5), with corresponding 95% limits of agreement of -12.6 to 10.5 HU and -39.1 to 37.5 HU, respectively. Density class analysis demonstrated that emphysematous areas were slightly underestimated (median, -0.4% [IQR, -1.4% to 0.1%]; P < .001) and normal-attenuation lung areas were modestly overestimated (median, +0.8% [IQR, -1.3% to 1.7%]; P < .001) on VNC images, without relevant changes in higher-density classes (both P ≥ .71). Voxel-wise differences were minimal within parenchyma (mean, 0.6 HU ± 15.1; P = .76) but more pronounced at anatomic interfaces. Conclusion In patients who underwent thoracic PCCT, VNC images demonstrated agreement with TNC scans for lung volume and attenuation, and inspiratory depth, rather than reconstruction method, was the dominant source of attenuation variability. © RSNA, 2026 Supplemental material is available for this article.