Room: ePoster Forums
Purpose: Despite decades of observational studies, central nervous system necrosis is a poorly understood radiation late effect. Ischemia and reperfusion caused by vascular injury is one of the proposed mechanisms behind its induction and progression. Computer simulations of vascular injury could provide new insights into this potentially fatal late effect. Simulations of radiation injury to vasculature and the resulting changes in blood flow are challenging because the human brain contains between 3 and 9 billion vessels with sizes from micrometers to centimeters. The goal of this project was to determine if it is feasible to perform first-principles biophysical simulations and quantify the minimum computational resources required.
Methods: We developed and implemented algorithms to construct a vascular geometry, simulate radiation dose, and predict the resulting changes in blood flow. We used a simple, fractal-like geometry and steady-state equations of blood flow. We utilized a recursive approach to multi-scaled geometry and dose scoring. The dose was calculated using an amorphous track structure model and stochastic track starting points. We measured execution time, memory use, and computational scalability of the algorithm on a 128-node supercomputer.
Results: The algorithms simulated the radiation dose from 2 million proton tracks to 4 billion vessels (human brain) in 49.3 hours. Peak memory use was less than 64 GB per node. A simulation of 1 billion protons incident on a mouse brain (4 million vessels) took 68.5 hours.
Conclusion: It is computationally feasible to simulate radiation dose and resulting changes in blood flow to all the blood vessels in the human brain. The algorithmâ€™s execution speed was limited by the geometric computations needed to calculate the dose. This first whole brain calculation suggests that new avenues of inquiry are possible. The availability of scalability data may help inform the design of future computational studies.
Funding Support, Disclosures, and Conflict of Interest: Funding for this work provided by the Bella Bowman Foundation and the US Nuclear Regulatory Commission.