Comparative Evaluation of Voxel-Level Dosimetry Methods in ⁹⁰ Y-Microsphere Radioembolization Using PET/CT

⁹⁰Y-microsphere radioembolization is an established treatment for primary and metastatic liver cancer. Post-treatment PET/CT enables voxel-level absorbed dose estimation, but the optimal calculation method remains under evaluation. In this study, we compared three dosimetry approaches: Monte Carlo (MC), voxel S-value (VSV) method, and local energy deposition method (LDM). Post-treatment PET/CT data sets from 68 patients were analyzed (6 from the Deep Blue Data Repository and 62 from Beijing Tsinghua Changgung Hospital). MC dosimetry was implemented in GATE v9.0 and served as the reference standard. VSV was derived by convolving PET images with a precomputed ⁹⁰Y voxel kernel, while LDM assumed complete local absorption of β-energy. Agreement with MC was assessed qualitatively using relative-difference maps, dose profiles, and isodose overlays, and quantitatively via voxelwise root-mean-square error (RMSE), Pearson correlation, Bland-Altman analysis of mean absorbed dose and equivalent uniform biological effective dose (EUBED), and absolute maximum deviations between cumulative and differential dose-volume histograms (DVHs) (1 Gy bins). MC simulations required ∼41 h per case, whereas the VSV kernel calculation required only one simulation with ∼6 h. Then VSV and LDM costed only a few seconds to obtain the dose images. VSV preserved MC isodose geometry and dose-profile shapes, while LDM overestimated doses in high-dose regions. At profile maxima, VSV's peak relative deviation was ∼1-3% versus ∼12-19% for LDM. Across patients, VSV achieved a voxelwise RMSE of 2.21% and <3% differences in mean dose and EUBED, with tighter Bland-Altman limits than LDM. LDM showed larger bias, wider limits, and greater DVH deviations at higher doses, confirming that VSV attains near-MC accuracy with orders-of-magnitude faster computation. Overall, all three methods enable voxel-level dosimetry of ⁹⁰Y-microsphere therapy, but VSV combines MC-like accuracy with dramatically reduced computation time. VSV thus offers a clinically practical solution for rapid and reliable post-treatment dose verification and may support more personalized treatment evaluation in ⁹⁰Y radioembolization.