Purpose: Ultra-high dose rate irradiation (FLASH) has emerged as a promising technique that may increase the therapeutic ratio of radiation therapy. A leading hypothesis is that the reduction of normal tissue toxicity is mediated by transient radioprotective hypoxia. A computational model is developed to simulate heterogeneous oxygen levels in the mouse brain during FLASH and investigate the effect of transient oxygen depletion on normal tissue complication probability.
Methods: A 3D model of mouse brain vasculature was constructed using a publicly available two-photon fluorescence microscopy dataset. A dynamic oxygen distribution was modeled by considering oxygen diffusion, metabolism and radiolytic depletion. During FLASH, the oxygen enhancement ratio (OER) was incorporated into the linear-quadratic (LQ) model to obtain cell killing rate and compute an effective dose distribution considering oxygen depletion and heterogeneous distribution.
Results: The effective dose is calculated for a FLASH pulse of 2-50Gy and mean oxygen levels 0.1-35mmHg in extra-vascular space. For a fixed physical dose, dose modifying factors (DMFs) first decrease and then increase with increasing oxygen tension. The oxygen tension at which DMF is minimized (maximal FLASH effects) shifts toward higher oxygen levels with increasing dose. For a fixed initial oxygen level, FLASH effects increase and then saturate with increasing physical dose. In the range 12-22Gy, V12 (a predictor of brain necrosis) yields a higher value for extreme cases of 0.1 and 35mmHg and a lower value for 6, 9, and 15mmHg.
Conclusion: The proposed computation model is a useful tool to analyze various factors, i.e., alpha, beta (in LQ), metabolism, oxygen tension, OER, physical dose affecting the FLASH effects. The tissue oxygen heterogeneity may play an important role in mediating the FLASH effect. The same concept can be easily applied to the tumor and eventually help provide insight into determining the optimal treatment parameters for the FLASH RT.
TH- Response Assessment: Modeling: other than machine learning