Personalized dosimetry for liver cancer radioembolization using fluid dynamics

Objectives: To improve individualized radioembolization (RE) treatment planning by developing new dosimetry calculation that incorporates patient data and fluid dynamics simulation. Liver cancer is one of the deadliest cancers worldwide [1]. RE is an efficient approach to treating liver cancer, in which yttrium 90 (⁹⁰Y) microspheres are injected into the liver through the hepatic artery to deliver localized radiation [2]. However current planning relies on standard internal dosimetry with poor accuracy for RE, which often results in suboptimal treatment. To address the limitations of the planning protocols, we propose to build a dosimetry tool for RE that integrates a computational fluid dynamics (CFD) model of the ⁹⁰Y microsphere distribution in the liver and ⁹⁰Y decay physics.

Methods: We tested our approach using ex vivo data obtained on rat livers. An X-ray opaque casting agent (Microfil, Flow Tech Inc) was injected post-mortem in the hepatic artery before scanning the liver at 30 µm spatial resolution with microCT (MicroXCT-200 Zeiss). The vasculature was segmented from the images, and for computational efficiency only a partial structure with six branches was retained for a proof-of-concept simulation (Fig. 1). The structure was then meshed with 660K cells and scaled to match the size of a human hepatic artery (~6 mm diameter)[3]. CFD simulation was conducted using the opensource software OpenFOAM with microspheres having the density of glass and a diameter of 30 µm, similar to TheraSphere (BTG Interventional Medicine), one of the two FDA-approved RE microspheres. A total of approximately 10,000 microspheres were introduced with the flow at the inlet of the vascular structure shown in Fig 1a. Although microspheres in the complete hepatic arterial tree would travel until they reach arterioles of comparable diameter, here termination of the CFD computational domain occurs at the six studied outlets shown in the figure, each labeled with the number of microspheres flowing across the respective outlet plane. The 3D distribution of the microspheres (i.e. located in the outlet planes) obtained from the CFD simulation was used to create a multi-foci source within a voxelized liver phantom for the dosimetry simulation. The geometry and properties of the microspheres, as well as the voxelized liver phantom and the brachytherapy physics, were simulated using the Geant4-based software for medical physics applications GAMOS[4]. Each microsphere was considered a ⁹⁰Y point source with an activity of 2500 Bq, which is the unit activity for glass microspheres. ⁹⁰Y is a β^- emitter with a 64h half-life, and a maximum energy of 2.28 MeV. The phantom properties were set as liver tissue of a healthy adult as described in ICRU Report 46.

Results: Figs. 1b-c show a voxelized histogram of the dose delivered to the liver tissue. The dose is summed for each 2D plane. The voxel size was 0.5 mm, indicating that high resolution dosimetry could be computed. As expected with the limited penetration range of ⁹⁰Y, the delivered dose is concentrated primarily around the locations of the microspheres; with a spread of 1-2 mm from the dimensions of the outlets. Doses are given in Gy/voxel on the colorbars and result from the 25 million events simulated in this proof-of-concept simulation.

Conclusion: We demonstrated that using an image-based vasculature mesh, voxelized dosimetry for RE could be achieved using a combination of simulations of CFD and ⁹⁰Y decay physics. Our ultimate goal is to apply this framework to human liver images acquired during standard-of-care RE planning. As both current internal dosimetry methods and ^99mTc-MAA SPECT that are used to plan the distribution of the microspheres have poor accuracy for RE, our work has the potential to improve treatment and outcome by providing a dosimetry calculation that can be tailored to each patient. Research Support: This work was supported by NIH grant R35 CA197608 and a UC Davis Innovation Developmental Award. $$graphic_518A1248-56B8-4E06-970F-B712786730BD$$