To compare the reliability and accuracy of the pre-treatment dosimetry predictions using cone-beam computed tomography (CBCT) versus Tc-labeled macroaggregated albumin (MAA) SPECT/CT for perfused volume segmentation in patients with hepatocellular carcinoma treated by selective internal radiation therapy (SIRT) using Y-glass microspheres. Fifteen patients (8 men, 7 women) with a mean age of 68.3±10.5 (SD) years (range: 47-82 years) who underwent a total of 17 SIRT procedures using Y-glass microspheres for unresectable hepatocellular carcinoma were retrospectively included. Pre-treatment dosimetry data were calculated from Tc-MAA SPECT/CT using either CBCT or Tc-MAA SPECT/CT to segment the perfused volumes. Post-treatment dosimetry data were calculated using Y imaging (SPECT/CT or PET/CT). The whole liver, non-tumoral liver, and tumor volumes were segmented on CT or MRI data. The mean absorbed doses of the tumor (D), non-tumoral liver, perfused liver (D) and perfused non-tumoral liver were calculated. Intra- and interobserver reliabilities were investigated by calculating Lin's concordant correlation coefficients (ρ values). The differences (biases) between pre- and post-treatment dosimetry data were assessed using the modified Bland-Altman method (for non-normally distributed variables), and systematic bias was evaluated using Passing-Bablok regression. The intra- and interobserver reliabilities were good-to-excellent (ρ: 0.80-0.99) for all measures using both methods. Compared with 90Y imaging, the median differences were 5.8Gy (IQR: -12.7; 16.1) and 5.6Gy (IQR: -13.6; 10.2) for D-CBCT and D-Tc-MAA SPECT/CT, respectively. The median differences were 1.6Gy (IQR: -29; 7.53) and 9.8Gy (IQR: -28.4; 19.9) for D-CBCT and D-Tc-MAA SPECT/CT respectively. Passing-Bablok regression analysis showed that both CBCT and Tc-MAA SPECT/CT had proportional biases and thus tendencies to overestimate D and D at higher post-treatment doses. CBCT may be a reliable segmentation method, but it does not significantly increase the accuracy of dose prediction compared with that of Tc-MAA SPECT/CT. At higher doses both methods tend to overestimate the doses to tumors and perfused livers.< Réduire