Program Information
Demonstration of a Concurrent Monte Carlo Dose Calculation and Fluence Optimization Platform with Low Memory Footprint
Y Yang1*, M Svatos2 , C Zankowski2 , B Bednarz1 , (1) University of Wisconsin, Madison, WI, (2) Varian Oncology Systems, Palo Alto, CA
Presentations
TH-EF-BRD-3 (Thursday, July 16, 2015) 1:00 PM - 2:50 PM Room: Ballroom D
Purpose:
To enable a treatment planning solution that utilizes voxel-specific penalties requires optimizing over a large solution space both quickly and accurately. Although challenging, such methods will enhance the clinical impact of adaptive (online and offline) and biologically guided radiotherapy. Monte Carlo calculations generally provide improved accuracy over deterministic methods, but with increased computation time. Instead, we present a low memory footprint direct Monte Carlo optimization platform that optimizes the fluence concurrently during the MC dose calculation. This reduces the wasted computation time spent on minimally contributing beamlets, and makes the combined optimization and dose calculation tasks both fast and accurate even over a large mathematical space.
Methods:
The Monte Carlo code Geant4 v9.6 was used to create the Fluence Adjusted Scaled Transport Monte Carlo (FASTMC) platform. The FASTMC platform minimizes clinical cost objectives during the dose calculation using multiple co-planar sources of arbitrary gantry angle and SID, for rotational or fixed beams. Beamlet particle histories are transported in iterative steps; cost is evaluated after each iteration, with beamlet intensity and dose profile incremented such that the cost is decreased. A Greedy Heuristic algorithm is employed that circumvents the necessity of a dose coefficient matrix, resulting in a memory footprint scaling that is independent of the patient CT resolution, and allows for intelligently biased selection, and calculation of beamlets based on costs in prior iterations.
Results:
A multi-field coplanar prostate plan was generated during the MC dose calculation, resulting in an optimized dose profile that is calculated concurrently with the optimized fluence. All clinical constraints for the target and OARs were met for this plan.
Conclusion:
This study demonstrated the feasibility of optimizing an incident fluence map using voxel-specific penalties during Monte Carlo dose calculations as a means of treatment planning tailored to emerging radiation therapy delivery paradigms.
Funding Support, Disclosures, and Conflict of Interest: This work is partially based upon an idea conceived by Varian Medical Systems. This work is partially supported by the NIH Training Grant 2T32CA009206-36A1.
Contact Email: