Abstract
2823
Introduction: Voxel-level radiopharmaceutical dosimetry offers the opportunity to individualize the dose administered to the patient, ensuring that a therapeutic level of radiation reaches the target lesions while ameliorating radiotoxicity in otherwise healthy tissues. Effective dosimetry requires quantitative radioisotope imaging; in SPECT this includes accurate compensation for photon attenuation, collimator blurring and Compton scatter. Different reconstruction algorithms, with different approaches to addressing these issues, will produce reconstructed images with different characteristics. This study was undertaken to investigate how those image differences are manifest in the computed dose distribution.
Methods: Longitudinal SPECT/CT datasets with lesion and organ contours were provided by the SNMMI Lu-177 Dosimetry Challenge (DC). The original SPECT images supplied by the DC were used in the study, together with images reconstructed using 10 iterations/10 subsets of MOSEM, an OS-EM algorithm where the system forward model is computed by Monte Carlo methods (Ryden et al., EJNMMI Physics (2018) 5:1). Voxel-level absorbed dose distributions were calculated using TorchTM, a GPU-based Monte Carlo planning system. The dose distributions were compared using dose statistics, DVHs, and dose difference maps, using organ and lesion ROIs supplied with the DC data.
Results: The mean dose to the large background ROI in the liver was 6-7% lower when calculated from the MOSEM images compared to the original DC images. The difference in mean dose increased to 13-21% in the smaller background lesions (kidneys, spleen) and lesions identified in the DC data sets. Evidence of Gibbs ringing at the edges of high-dose lesions could be found in the plan computed from the original DC images, consistent with the application of resolution recovery algorithms in those reconstructions; ringing was not seen in those locations in the plan computed from the MOSEM reconstructions.
Conclusions: The reconstruction methods used in preparing images for radiopharmaceutical dosimetry calculations have a substantial impact on the resulting voxel-level dose distributions which may be clinically relevant.