An evaluation of dual-layer detector spectral CT for the estimation of MRI proton density fat fraction of the liver.

Background: Magnetic resonance imaging (MRI) proton density fat fraction (PDFF) is the preferred method for investigating steatotic liver disease. Recent studies with dual-layer computed tomography (DLCT) suggest that it might provide an alternative imaging modality to MRI PDFF. The aim of this study was to determine which DLCT scan parameter provides the most reliable estimate of liver PDFF. The second aim was to show that the best parameter is the one that optimally combines attenuation data from both the photoelectric effect and Compton scattering. Methods: This prospective study included ninety-nine healthy community-dwelling adults who underwent a non-contrast upper abdominal CT examination using a 64-slice DLCT system and a same-day Q-Dixon study with a 3.0-T MRI scanner. Effective atomic number (Zeff), electron density (ED), and Hounsfield units (HU) of the poly-energy CT image and virtual mono-energetic images at 17 energies between 40 and 200 keV were measured and compared with MRI PDFF using the Pearson correlation coefficient, the standard error of the estimate (SEE), and receiver operating characteristic (ROC) analysis. A parallel technical study was undertaken using the Zeff and ED data together with National Institute of Standards and Technology (NIST) X-ray attenuation coefficients to estimate the relative contributions of the photoelectric effect and Compton scattering to CT liver fat measurements at energies between 40 and 200 keV. Results: Mono-energy HU measurements at 50 keV provided the highest correlation coefficient, lowest SEE, and highest ROC area under the curve (AUC) (r=0.948; SEE =1.87%, AUC =0.940). Results for mono-energy measurements at 60 and 70 keV and poly-energy measurements were not significantly different. The two worst-performing measurements were Zeff (r=0.849, SEE =3.11%, AUC =0.871) and ED (r=0.902, SEE =2.55%, AUC =0.868), which were both statistically significantly poorer at predicting MRI PDFF than the 50 keV HU measurement (P<0.001). The study using NIST data showed that the contribution of the photoelectric effect to liver fat measurements varied from 65% at 40 keV to 5% at 200 keV and was approximately 50% at 50 keV. Conclusions: Zeff and ED measurements are least effective at estimating liver fat because they exclude information from Compton scattering and the photoelectric effect, respectively. HU measurements between 50 and 70 keV improve the detection of liver steatosis by utilising information from both processes in an optimum ratio of approximately 50:50.