A noise-optimized virtual monoenergetic reconstruction algorithm improves the diagnostic accuracy of late hepatic arterial phase dual-energy CT for the detection of hypervascular liver lesions

Objectives To assess the image quality and diagnostic accuracy of a noise-optimized virtual monoenergetic imaging (VMI+) algorithm compared with standard virtual monoenergetic imaging (VMI) and linearly-blended (M_0.6) reconstructions for the detection of hypervascular liver lesions in dual-energy CT (DECT). Methods Thirty patients who underwent clinical liver MRI were prospectively enrolled. Within 60 days of MRI, arterial phase DECT images were acquired on a third-generation dual-source CT and reconstructed with M_0.6, VMI and VMI+ algorithms from 40 to 100 keV in 5-keV intervals. Liver parenchyma and lesion contrast-to-noise-ratios (CNR) were calculated. Two radiologists assessed image quality. Lesion sensitivity, specificity and area under the receiver operating characteristic curves (AUCs) were calculated for the three algorithms with MRI as the reference standard. Results VMI+ datasets from 40 to 60 keV provided the highest liver parenchyma and lesion CNR ( p ≤0.021); 50 keV VMI+ provided the highest subjective image quality (4.40±0.54), significantly higher compared to VMI and M_0.6 (all p