[Paper Review] Phantom study for 90Y post-treatment dosimetry with a long axial field-of-view PET/CT

Lorenzo Mercolli, Konstantinos Zeimpekis+4 more|arXiv (Cornell University)|Apr 28, 2023

Medical Imaging Techniques and ApplicationsMedicine3 citations

TL;DR

This study evaluates post-treatment dosimetry for 90Y liver radioembolization using long axial field-of-view (LAFOV) PET/CT, comparing Monte Carlo simulations, Simplicit90Y (image-independent), and Hermes (image-based sMC algorithm). It finds that Hermes underestimates dose due to poor PET quantification in small structures, while Simplicit90Y and Monte Carlo agree well, highlighting limitations in image-based dosimetry despite improved LAFOV sensitivity.

ABSTRACT

Purpose: The physical properties of yttrium-90 (90Y) allow for imaging with positron emission tomography/computed tomography (PET/CT). The increased sensitivity of long axial field-of-view (LAFOV) PET/CT scanners possibly allows to overcome the small branching ratio for positron production from 90Y decays and to improve for the post-treatment dosimetry of 90Y of selective internal radiation therapy. Methods: For the challenging case of an image quality body phantom, we compare a full Monte Carlo (MC) dose calculation with the results from the two commercial software packages Simplicit90Y and Hermes. The voxel dosimetry module of Hermes relies on the 90Y images taken with a LAFOV PET/CT, while the MC and Simplicit90Y dose calculations are image independent. Results: The resulting doses from the MC calculation and Simplicit90Y agree well within the error margins. The image-based dose calculation with Hermes, however, consistently underestimates the dose. This is due to the mismatch of the activity distribution in the PET images and the size of the volume of interest. Furthermore, there are likely limitations of Hermes' dose calculation algorithm for 90Y. We found that only for the smallest phantom sphere there is a statistically significant dependence of the Hermes dose on the image reconstruction parameters and scan time. Conclusion: Our study shows that Simplicit90Y's local deposition model can provide a reliable dose estimate. On the other hand, the image based dose calculation requires further benchmarks and verification in order to take full advantage of LAFOV PET/CT systems.

Motivation & Objective

• To assess the accuracy of post-treatment 90Y dosimetry using long axial field-of-view (LAFOV) PET/CT in a phantom study.
• To compare image-independent dosimetry methods (Monte Carlo and Simplicit90Y) with image-based dosimetry using the Hermes software.
• To investigate whether LAFOV PET/CT improves 90Y dosimetry accuracy despite the low branching ratio for positron emission.
• To identify the primary source of dose underestimation in Hermes software—poor activity quantification or algorithmic limitations.

Proposed method

• Conducted a phantom study using two NEMA IEC body phantoms with 90Y citrate in spheres and background, measuring 1.12 ± 0.17 GBq at scan time.
• Acquired LAFOV PET/CT images at multiple scan times (50 min UHS, 20 min HS, 10 min HS) to assess image quality impact on dosimetry.
• Performed full Monte Carlo dose calculations using FLUKA to serve as ground truth, modeling 90Y decay, particle transport, and dose deposition independently of imaging.
• Used Simplicit90Y’s local deposition model for dose calculation, independent of PET images, assuming uniform β⁻ energy deposition.
• Applied Hermes’ semi-Monte Carlo (sMC) algorithm using reconstructed PET images for dose mapping, assessing sensitivity to reconstruction parameters and scan time.
• Generated synthetic PET images to isolate the effect of image quantification from algorithmic limitations in Hermes.

Experimental results

Research questions

• RQ1 Does LAFOV PET/CT improve the accuracy of 90Y post-treatment dosimetry compared to conventional systems?
• RQ2 Why does Hermes software systematically underestimate 90Y dose compared to Monte Carlo simulations?
• RQ3 To what extent do image reconstruction parameters and scan time affect Hermes’ dose estimates in small structures?
• RQ4 How do image-independent methods like Simplicit90Y and Monte Carlo compare in their dose predictions for 90Y microspheres?
• RQ5 Is the underestimation in Hermes due to limitations in the sMC algorithm or poor quantification of 90Y activity in PET images?

Key findings

• The Monte Carlo and Simplicit90Y dose calculations agree well within error margins, confirming Simplicit90Y’s reliability for larger volumes.
• Hermes consistently underestimates dose across all phantoms, with the largest discrepancies in the smallest sphere (diameter 10 mm), indicating poor quantification in small structures.
• The underestimation in Hermes is primarily due to inaccurate activity quantification in PET images rather than inherent flaws in the sMC algorithm, as shown by synthetic image analysis.
• Even with longer scan times (50 min UHS), Hermes dose estimates do not significantly improve, and the trend is not statistically significant due to large error bars.
• The sMC algorithm overestimates dose in small spheres when applied to synthetic images, but this effect is overshadowed by image quantification errors in real data.
• Recovery coefficients in LAFOV PET/CT do not reach 100%, even for the largest sphere, confirming limitations in quantifying small-volume activity distributions.