Retrospective PET/CT Imaged Based Dosimetry Following Transarterial Radioembolization

Abstract

Background: Transarterial Radioembolization (TARE) uses beta particle emitting microspheres, containing Yttrium 90 (90Y), to treat unresectable primary and metastatic liver tumors. Current dosimetry models are highly simplistic. 90Y post-TARE PET/CT images are being evaluated for their use in image based dosimetry post treatment allowing for improved treatment efficacy. We sought to provide guidance on reconstruction and dose calculation algorithms that give the most reliable image based dosimetry.Methods: Retrospective image study for 11 patients each having a 90Y PET/CT following TARE SIR-Spheres® treatment. In total, 6 emission images were reconstructed using Ordered Subset Expectation Maximization (OSEM) routines and Bayesian penalized-likelihood reconstruction algorithms (Q. Clear). PET/CT image sets were resampled to 3mm resolution and used to create voxel based dose distributions using our own convolution algorithm (WFBH), SurePlanTM MIRD and SurePlanTM LDM. Additionally, we analyzed the effects of a systematic personalized background subtraction prior to dose distribution creation. Results: Reconstructed activity was highest among Q. Clear β = 350 and 1000 and both OSEM methods. OSEM with 3 iterations and 24 subsets gave comparable dose distributions to Q. Clear β=1000 (max dose ratio of 0.96 ± 0.14). No statistical difference was identified among Q. Clear and OSEM methods comparing dose distributions with and without additional background subtraction post-reconstruction (average local gamma value = 98.01% ± 3.75%). LDM is more sensitive than MIRD to small activity differences between reconstruction methods (p-value= 0.048). LDM calculated max dose values were higher than MIRD. DVH curves showed LDM model potentially underestimates low dose values and overestimates high dose values.Conclusion: Background subtraction is sufficiently addressed in PET image reconstruction. We recommend post-TARE image based dosimetry to be calculated using image series reconstructed with Q. Clear β= 1000 or OSEM 3 iterations 24 subsets and using a convolution voxel based dose algorithm, such as SurePlanTM MIRD. Since more sophisticated corrections are not available at present, the 90Y activity distribution should be scaled by the local clinic’s activity recovery fraction, which also will directly scale the calculated dose distribution.