LAUSR

PURPOSE To investigate the association of vertebral CT attenuation between virtual non-contrast (VNC) and true non-contrast (TNC) images and to evaluate if VNC vertebral CT attenuation could be used for phantom-less osteoporosis detection in dual-layer spectral-detector CT (SDCT). METHODS 200 patients with non-contrast and portal-venous phase SDCT were retrospectively assigned to a test and a validation group of 100 patients each. CT attenuation of L1 vertebrae were measured on VNC and TNC. The test group was used to determine the difference between VNC and TNC CT attenuation and to calculate a statistical model for TNC CT attenuation prediction. The validation group was used to assess the capability of the model to predict TNC from VNC CT attenuation and its accuracy to identify osteoporosis. Osteoporosis was defined as a TNC CT attenuation of ≤110HU. RESULTS In both groups, CT attenuation was lower in VNC than in TNC (P < 0.001). VNC and TNC CT attenuation was correlated strongly (r = 0.958). Using the regression equation established in the test group (TNC = 23.677 + 1.540 × VNC), the predicted TNC CT attenuation did not differ from the real TNC CT attenuation in the validation group (P = 0.359). A VNC CT attenuation cut-off of 52HU yielded an AUC of 0.978 for osteoporosis detection. CONCLUSIONS L1 CT attenuation is systematically underestimated in VNC compared with TNC images. However, TNC L1 CT attenuation can be predicted reliably from VNC. VNC may perform well in phantom-less osteoporosis detection.