Abstract
P741
Introduction: Quantitative PET can estimate in vivo 90Y-activity and absorbed dose distributions following 90Y-radioembolization (90Y-RE). Due to its low β+ branching ratio (32 ppm), 90Y-PET acquisitions are long (10-20 min/bed) and subject to respiratory motion blur and image coregistration errors, whose impact on the accuracy of 90Y-activity quantification are largely unknown. Furthermore, absorbed doses are estimated in PET by dose kernel-based voxel-S-values (VSV) dosimetry in lieu of local deposition model with known activity (LDM). In this work, we characterize the accuracy of in vivo 90Y-activity estimates (Aimage) relative to administered 90Y-activity (Aadmin) in perfused volumes (PV) and tumor (T) volumes of interest (VOIs). We also investigate differences between DVSV (mean dose from VSV dosimetry) and DLDM (mean dose from LDM dosimetry) in PV and T.
Methods: As part of a prospective 90Y-RE clinical trial RAPY90D (NCT03896646) using TheraSphere, 33 patients underwent 90Y-PET/CT imaging (GE Discovery MI-5ring) using 90Y-radionuclide selection for a single bed 25-minute acquisition. Images were reconstructed with CT-attenuation and scatter corrections using 3D-OSEM (2 IT, 21 SUB) in 2.7x2.7x2.8 mm3 voxels. PV (N=42), PV with 2 cm isotropic expansion (PV2) and T (N=47) VOIs were delineated by an interventional radiologist and used in conjunction with MIM SurePlan LiverY90 for dosimetry and 90Y-activity quantification. Aimage in PV and PV2 were compared to Aadmin using Bland-Altman (BA) analysis, with and without patient-specific lung shunt fraction (LSF) correction. For each comparison, we report correlations, mean bias, standard error (SE), limits of agreement (LOA), and fraction of cases with ±10% and ±20% differences. The ratio of Aadmin to planned activity (Aplan) is also reported. Likewise, we calculated mean bias, SE, and LOA in BA comparisons of DLDM and DVSV in PV and T. Finally, we investigated potential correlations between activity differences for PV (DAPV) and Aadmin and volume, as well as correlations between dose differences for PV and T (DDPV and DDT) and DLDM and volume. In each comparison, Aadmin and DLDM served as the ground truth.
Results: The median (IQR) Aadmin/Aplan was 1.006 (0.007) demonstrating high precision for Aadmin. Although Aadmin and Aimage were correlated (R2>0.93), the mean bias±SE (LOA) between Aimage and Aadmin in PV and PV2 were -33±5% (±32%) and 1±8% (±51%), respectively. A paired t-test showed significant differences between Aadmin andAimage for PV (0.66±0.14 GBq, p<0.01) but not for PV2 (0.08±0.16 GBq, p=0.57). The fraction of cases with ±10%, ±20% differences between Aimage and Aadmin increased from 5% and 19% for PV to 48% and 67% for PV2, respectively. With a median (IQR) LSF of 3.2% (2.9%), there were no significant differences in the mean bias and residual error distribution when Aadmin was adjusted by LSF. DAPV decreased linearly with Aadmin (p<0.01) and with PV volume (p<0.01). In terms of dosimetry for PV and T, DLDM and DVSV were well correlated (R2=0.70 and 0.75) but resulted in a mean bias±SE (LOA) of 23±6% (±36%) and 23±5% (±33%), respectively. Both DDPV and DDT were weakly correlated with DLDM (R2=0.54 and 0.34), but only DDT was correlated with T volume (R2=0.23).
Conclusions: While Aimage was highly correlated with Aadmin for PV and PV2, yet only by accounting for "activity blurring" from respiratory motion and coregistration errors (PV2) wasAimage, on average, found to equal Aadmin. However, 90Y-activity agreement on an individual patient basis (precision) was low with LOA of ±51%. These findings were not affected by LSF correction and Aplan assay variability. The mean bias between Aadmin and Aimage and between DLDM and DVSV was 33% and 23%, respectively, and independent of VOI assessed. The 10% discrepancy observed may be attributed to differences in the specific implementation of VSV dosimetry or differences in TheraSphere activity calibration. Ongoing work is focused on understanding these discrepancies.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.