Photon counting detector CT-derived virtual non-contrast images of the liver: comparison of conventional and liver-specific algorithms across arterial and portal venous phase scans

The aim of this retrospective study is to compare photon-counting detector computed tomography (PCD-CT) derived virtual non-contrast (VNC) images of the liver reconstructed from both arterial and portal venous phase using conventional and liver-specific VNC algorithm to true non-contrast images, in context of the body mass index (BMI). VNC images reconstructed from multiphase (non-contrast, arterial and portal venous phase) PCD-CT scans performed between April 2021 and February 2023 were analysed retrospectively. For each patient, four VNC series were generated: two series (arterial and portal venous) using a conventional VNC algorithm (VNCconvart; VNCconvpv) and two using a liver-specific “Liver VNC” algorithm (VNCLiverart; VNCLiverpv). Regions of interest were placed in the left and right liver lobes and in the spleen, avoiding large vessels and focal lesions. The VNC CT-values were then compared to those of the corresponding true non-contrast images (TNC). The subsequent analysis involved the calculation of both correlation and mean offsets. The median split was utilised to ascertain distinct cohorts of patients with elevated and reduced body mass indices. These cohorts were then subjected to a comparative analysis of attenuation values to discern potential disparities between them. The results were compared by using parametric and non-parametric tests; Pearson’s correlation coefficient was employed. Bland-Altman plots were utilised to visually assess the agreement between results and Passing-Bablok regression, thereby quantifying the observed agreement. The study population comprised 42 patients (mean age 70.0 ± 10.2 years, 33 males). Mean offsets between TNC and VNCconvart was 0.62 ± 5.23 HU, TNC-VNCconvpv 1.24 ± 6.67 HU, TNC-VNCLiverart -0.94 ± 5.59 and TNC-VNCLiverpv -0.35 ± 6.99 with no significant difference. Significant differences were found for VNCconvart, VNCconvpv and VNCLiverart images regarding spleen attenuation. Bland-Altman plots demonstrated good agreement and the absence of any systematic difference in liver attenuation. As for the TNC-VNCconvart, TNC-VNCconvpv, TNC-VNCLiverart and TNC-VNCLiverpv variables, strong correlations were obtained (Pearson’s coefficient: 0.79, 0.69, 0.79 and 0.7, all p < 0.001). The investigation revealed no statistically significant disparities between the BMI groups with respect to the mean offset of liver density (p-value:TNC-VNCconvart 0.51; VNCconvpv 0.61; VNCLiverart 0.68; VNCLiverpv 0.45). Furthermore, no significant offset between TNC and VNC images was detected within each BMI group. A Passing-Bablok regression analysis revealed no systematic or proportional difference between the two methods. It is evident that PCD-CT-derived VNC images generally constitute a corresponding alternative to TNC images. However, caution is advised in the interpretation of images, as there are outliers with differences exceeding 15 HU are present. In general, the mean values obtained from the analysis of, VNC images reconstructed from arterial and portal venous phases employing both the liver-specific and general VNC reconstruction algorithm did not demonstrate any clincially significant difference when compared with TNC images. Furthermore, no significant discrepancy was observed in the utilisation of the conventional and the liver-specific algorithm. The findings of this study demonstrated that, within the limitations of the study, the patients’ BMI did not have a significant impact on the VNC images.