Differentiation of benign and malignant lung lesions: Dual-Energy Computed Tomography findings

Abstract

PurposeTo determine whether parameters generated by Dual-Energy Computed Tomography (DECT) can distinguish malignant from benign lung lesions. MethodsA prospective review of 125 patients with 126 lung lesions (23 benign and 103 malignant) who underwent lung DECT during arterial phase. All lesions were confirmed by tissue sampling. A radiologist semi-automatically contoured lesions and placed regions of interest (ROIs) in paravertebral muscle (PVM) for normalization. Variables related to absorption in Hounsfield units (HU), effective atomic number (Zeff), iodine concentration (ρI) and spectral CT curves were assessed. Receiver operating characteristic (ROC) curves were generated to calculate sensitivity and specificity as predictors of malignancy. Multivariate logistic regression analysis was performed. ResultsReproducibility of measures normalized with PVM was poor. Bivariate analysis showed minimum Zeff and normalized mean Zeff to be statistically significant (p=0.001), with area under the curve (AUC) values: 0.66 (CI 95% 0.54–0.80) and 0.72 (CI 95%, 0.60–0.84), respectively. Logistic regression models showed no differences between raw and normalized measurements. In both models, minimum HU (OR: 0.9) and size (OR: 0.1) were predictive of benign lesions. ConclusionsA quantitative approach to DECT using raw measurements is simpler than logistic regression models. Normalization to PVM was not clinically reliable due to its poor reproducibility. Further studies are needed to confirm our findings.