Improved realism of hybrid mouse models may not be sufficient to generate reference dosimetric data

Recent developments of hybrid realistic models, such as Moby (mouse) and Roby (rat) developed by Segars et al. ["Development of a 4-D digital mouse phantom for molecular imaging research," Mol. Imaging Biol. 6, 149-159 (2004)] have found several applications in preclinical experiments. Indeed, their improved realism and flexibility in terms of mass scaling represent an attractive option for absorbed dose calculations based on "representative" models. However, the range of radiations involved in small animal molecular imaging and radiotherapy is of the same order of magnitude as organs of interest dimensions. As a consequence, minor geometric variations between rodents may lead to major differences in absorbed dose calculations. This study aims at validating a voxel-based model for use in absorbed dose estimates with two Monte Carlo codes and at assessing the dosimetric impact of Moby-based models definition.

The authors generated a 30 g-mouse phantom based on realistic hybrid model Moby (version 1). Dosimetric calculations (S-values, specific absorbed fraction) were performed with two Monte Carlo codes (MCNPX v2.7a and GATE v6.1) for (18)F, and a comparison with values published for Radiation Dose Assessment Resource realistic animal series was made. Several parameters such as material definition∕densities, fine suborgan segmentation for airways (trachea, lungs, remaining body), bones (ribs, spine, skull, remaining bones), heart (blood pool and myocardium), and stomach (wall and gastrointestinal content) were further studied, as well as nuclear data and spatial sampling.

Most organ masses matched the reference model (Moby v1) within ± 6%, except lungs, thyroid, and bones for which differences could reach 29%. Comparison of S-values (especially self-S-values) was consistent with mass differences observed between the two models. The reciprocity theorem for source∕target pairs was satisfied within few percents for specific absorbed fractions (g(-1)). However, significant discrepancies, reaching 160%, were observed for mutual liver∕stomach∕spleen S-values and could not be directly related to mass variations. Nonetheless, differences between S-values calculated with MCNPX and GATE for our model remained in the order of a few percents, i.e., within statistical uncertainties. Besides, modifications of organ densities increased S-values up to a factor 50 for the lungs∕thyroid pair when upper airway was properly segmented out of the body. Specific material composition and densities for several bone types led to a 10% decrease of S-values from the bone source to several target organs. Moreover, relative differences up to 100% were observed for S(stomach wall⇐spleen) when improving spatial-sampling by a factor 3.

This study demonstrated that comparison between two "similar" realistic digital mouse whole-body phantoms generated from the same software still led to very different S-values, even when total body and organ mass scaling were performed. Moreover, parameters such as organ segmentation, tissue material∕density, or spatial sampling should be defined and reported with great care to perform accurate small animal absorbed dose calculation based on "reference" models.