A comparison of methods for in vivo activity and absorbed dose quantification with PET/CT following yttrium‐90 radioembolization

Background Yttrium‐90 ( 90 Y) positron emission tomography (PET)/computed tomography (CT) imaging is increasingly being used to perform tumor (T) and normal liver (NL) voxel dosimetry after 90 Y‐radioembolization ( 90 Y‐RE). Yet, the accuracy of in vivo 90 Y‐PET/CT imaging, subject to motion blur and co‐registration inaccuracies, and 90 Y‐PET/CT dose quantification, subject to availability of different voxel dosimetry algorithms, are not well understood. Purpose The purpose of this study was to investigate the accuracy of 90 Y‐PET/CT‐based activity estimates following 90 Y‐RE and characterize differences between 90 Y‐PET/CT‐based voxel dosimetry algorithms. Methods Thirty‐five patients underwent 90 Y‐PET/CT imaging after 90 Y‐RE with TheraSphere. The net administered 90 Y activity (A admin ) was determined using a dose calibrator and pre‐ and post‐procedure exposure rate measurements. The summation of image‐based activity (A image ) was extracted from perfused volume (PV) and 3D‐isotropically 2‐cm expanded PV contour (PV +2 cm ). Absorbed doses were calculated using voxel S‐value (VSV), local deposition method (LDM), and LDM with known activity (LDM KA ) dosimetry algorithms. Linear regression and Bland‐Altman analysis quantified the relationship between A image and A admin and between mean dose estimates (D LDM , D VSV , D LDM‐KA ) for PV, T, and perfused NL volumes. Results While A admin and A image in PV were highly correlated ( R 2 > 0.95), the mean bias ± standard error (SE) and (95% limits of agreement, LOA) was significantly non‐zero with −22.7 ± 4.7% (± 28.4%). In PV +2 cm , the mean bias ± SE (± LOA) decreased to 1.3 ± 3.4% (± 18.0%) consistent with zero mean error. D LDM and D VSV were highly correlated ( R 2 > 0.99) for all volumes of interest (VOIs) and the mean bias ± SE (± LOA) was 2.2 ± 0.2% (± 1.0%), 0.7 ± 0.4% (± 2.8%), and 3.2 ± 0.5% (± 2.8%) for PV, T, and NL, respectively. D LDM‐KA and D VSV were correlated with R 2 = 0.86, 0.80, and 0.86 for PV, T, and NL, respectively. The mean bias ± SE (± LOA) between D LDM‐KA and D VSV was significantly non‐zero with −19.6 ± 5.1% (± 31.0%), −20.8 ± 4.4% (± 29.0%), and −18.1 ± 5.3% (± 31.1%) for PV, T, and NL, respectively. Conclusions The summation of A image in PV was underestimated relative to A admin . Only by accounting for respiratory motion, limited spatial resolution, and PET/CT co‐registration errors through VOI expansion was A image , on average, equal to A admin . The differences between D LDM and D VSV were not clinically relevant, though D LDM‐KA was approximately 20% greater than D VSV . Given the high quantitative accuracy of dose calibrators and challenges associated with accurate 90 Y‐PET/CT quantification, LDM KA is the preferred algorithm for accurate 90 Y‐PET/CT‐based dosimetry following 90 Y‐RE.