Accuracy and reproducibility of a cone beam CT-based virtual parenchymal perfusion algorithm in the prediction of SPECT/CT anatomical and volumetric results during the planification of radioembolization for HCC

Objectives To evaluate anatomical and volumetric predictability of a cone beam computed tomography (CBCT)-based virtual parenchymal perfusion (VPP) software for the single-photon-emission computed tomography (SPECT)/CT imaging results during the work-up for transarterial radioembolization (TARE) procedure in patients with hepatocellular carcinoma (HCC). Methods VPP was evaluated retrospectively on CBCT data of patients treated by TARE for HCC. 99m Tc macroaggregated albumin particles ( 99m Tc-MAA) uptake territories on work-up SPECT/CT was used as ground truth for the evaluation. Semi-quantitative evaluation consisted of the ranking of visual consistency of the parenchymal enhancement and portal vein tumoral involvement on VPP and 99m Tc-MAA SPECT/CT, using a three-rank scale and two-rank scale, respectively. Inter-reader agreement was evaluated using a kappa coefficient. Quantitative evaluation included absolute volume error calculation and Pearson correlation between volumes enhanced territories on VPP and 99m Tc-MAA SPECT/CT. Results Fifty-two CBCTs were performed in 33 included patients. Semi-quantitative evaluation showed a good concordance between actual 99m Tc-MAA uptake and the virtual enhanced territories in 73% and 75% of cases; a mild concordance in 12% and 10% and a poor concordance in 15%, for the two readers. Kappa coefficient was 0.86. Portal vein involvement evaluation showed a good concordance in 58.3% and 66.7% for the two readers, respectively, with a kappa coefficient of 0.82. Quantitative evaluation showed a volume error of 0.46 ± 0.78 mL [0.01–3.55], and Pearson R 2 factor at 0.75 with a p value