Image Fusion in Dual Energy Computed Tomography for Detection of Hypervascular Liver Hepatocellular Carcinoma Phantom and Preliminary Studies

Objective: This study was designed to determine the optimal blending method and parameters to fuse computed tomography (CT) data sets with different energy levels in dual-energy CT (DECT) for the detection of hypervascular liver lesions. Materials and Methods: A liver agar phantom containing 8 conical tubes with various concentrations of contrast material, was scanned using a Somatom Definition Dual Source CT (DSCT; Siemens, Forchheim, Germany) scanner in the dual energy mode at different current settings. CT data sets obtained at voltage potentials of 80 kVp and 140 kVp were fused using the linear blending method and nonlinear method with different weighting factors (0.1, 0.3, 0.5, 0.7, and 0.9) and different parameters sets (A-lambda: 20, omega: 430; B-lambda: 20, omega: 70; C-lambda: 250, omega: 430; D-lambda: 250, omega: 70). In 20 patients with hepatocellular carcinomas, multiphasic liver CT scans including arterial, portal, and equilibrium phases were performed. DECT was used only during the arterial phase but a voltage potential of 120 kVp was used for both the portal and equilibrium phases. For quantitative analyses of the phantom and patient study, the contrast-to-noise ratio (CNR) of the lesion to liver on arterial phase images, was measured. For qualitative analysis of the CT images of the 20 study patients, 5 radiologists, each with a different level of clinical experience, independently assessed the 5 types of image sets regarding lesion conspicuity and overall image quality. This study followed the guidelines of our hospital's institutional review board, and patient informed written consent was not required. Statistical comparisons were made using repeated measures ANOVA with Bonferroni correction for multiple comparisons. Results: For the phantom and patient studies, 2 linear images with weighting factors 0.5 and 0.7 and 2 nonlinear images with a wide width, showed a higher CNR of hyperattenuated lesions than a standard 0.3 weighting factor linear blended image (P < 0.05). For the patient study, a weighting factor 0.5 and a 0.7 linear mixing image had a 9.2% and an 11.8% increase in CNR, respectively, more than a 0.3 linear blended image (P < 0.05). Moreover, a nonlinear mixing image with parameters A and C had a 14.0% increase in CNR over that of a 0.3 linear blended image (P < 0.05). In a qualitative study performed by the 5 reviewers, the nonlinear blended image set with a low level and wide width, was estimated as the most preferred image set (55%-100%), whereas a weighting factor of a 0.3 linear blended image was determined as the least preferred image (65%-100%). Conclusions: Linear blended images with a higher weighting factor than 0.5 and nonlinear blended images with a wide width are able to provide improved lesion-to-liver CNR over the standard linear blending method (with a weighting factor of 0.3). Furthermore, the linear blending method with a low level and a high width provided the most preferred image set for hypervascular hepatocellular carcinoma detection by our radiologists.