Quantitative comparison of virtual non-contrast and true non-contrast computed tomography pulmonary angiography images across eight thoracic tissue types: A retrospective cross-sectional study.

ObjectiveTo evaluate the quantitative reliability of virtual non-contrast images derived from spectral computed tomography pulmonary angiography across eight thoracic tissues, assessing their feasibility as a true non-contrast substitute.MethodsIn this retrospective cross-sectional study, we analyzed 169 patients undergoing spectral computed tomography pulmonary angiography. Attenuation differences between true non-contrast and virtual non-contrast images were evaluated across eight anatomical regions. Statistical assessments included Bonferroni-corrected Wilcoxon signed-rank tests, Pearson correlation (r), Lin's concordance correlation coefficient, and Bland-Altman analysis.ResultsVirtual non-contrast performance showed four distinct patterns. (1) Excellent agreement. Lung parenchyma and erector spinae muscle demonstrated high absolute concordance (concordance correlation coefficient = 0.981 and 0.876, respectively) with clinically negligible biases (3.8 Hounsfield units and -1.0 Hounsfield units). (2) Moderate agreement. The aorta and pulmonary artery shared identical concordance correlation coefficient values (0.430) but exhibited divergent clinical threshold exceedances, primarily driven by a larger mean bias in the pulmonary artery (8.8 Hounsfield units). (3) Stable systematic offsets. The T6 vertebral body and mediastinal fat exhibited massive systematic biases (66.2 Hounsfield units and -11.1 Hounsfield units, respectively), leading to poor absolute agreement (concordance correlation coefficient = 0.390 and 0.531). However, both maintained strong linear correlations (r = 0.902 and 0.816) and extremely low outlier rates beyond the 95% limits of agreement. (4) Unacceptable agreement. The thyroid gland and superior vena cava yielded clinically unacceptable reliability due to proportional bias and beam-hardening artifacts. The physical true non-contrast acquisition accounted for 53.0% of the total effective radiation dose.ConclusionsSpectral virtual non-contrast imaging accurately substitutes true non-contrast imaging in the lung parenchyma and erector spinae muscle. Although major vessels show moderate reliability, the massive systematic underestimation in vertebral bone precludes its direct clinical use for quantitative assessment without validated mathematical correction factors. Intrinsically iodine-rich organs and structures prone to streak artifacts remain unsuitable for virtual non-contrast substitution.