Program Information
Towards a Multiscale Model of Vascular Dose for the Whole Brain
W Donahue1*, W Newhauser1,2 , (1) Louisiana State University, Baton Rouge, LA (2) Mary Bird Perkins Cancer Center, Baton Rouge, LA
Presentations
SU-I-GPD-T-654 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To develop a multiscale framework for vascular dose calculations and to determine the computational requirements of performing those calculations for the number of vessels in the human brain.
Methods: We developed novel methods to generate a fractal-based network of 4 billion vessels, matching the number in the human brain. We prototyped the network generation algorithm in MATLAB and ported it to C++ language for parallel computing. A recursive approach was taken to calculate proton dose to the vascular network on multiple dimensional scales. We prototyped a multiscale dose algorithm using the straight-ahead and continuous slowing down approximations to transport protons axially through the network, and an amorphous track structure model to compute dose radially. We separately characterized the computational performance of the codes for vascular creation and dose calculation for various network sizes.
Results: Generating a 4 billion vessel network took 42 CPU-hours, of which 37 CPU-hours was spent performing I/O operations. The MATLAB prototype of the vascular network generator was 737 times slower than the C program. Storing the network and all the necessary descriptors on disk required 409 GB. The multiscale dose algorithm required 15 CPU-hours to calculate the dose from 1,000 200-MeV protons incident on a 132,000 vessel network. Extrapolating to 1 billion particles and 4 billion vessels, accounting for the performance gains seen when porting the vessel constructor, revealed only an additional 60-fold speedup will allow 4096 CPUs to perform the dose computations in 3 days.
Conclusion: This study demonstrated that the algorithmic generation of a simple vascular network for the whole brain is entirely achievable. Secondarily, we obtained promising preliminary results for calculating multiscale dose to the whole-brain for use in research applications.
Funding Support, Disclosures, and Conflict of Interest: This work was funded by a grant provided by the Bella Bowman Foundation.
Contact Email: