Premium
SU‐GG‐T‐604: GPU‐Accelerated KV/MV Dose Computation
Author(s) -
Jacques R,
Taylor R,
Wong J,
McNutt T
Publication year - 2010
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.3469005
Subject(s) - superposition principle , computation , attenuation , collimator , computer science , kernel (algebra) , physics , algorithm , optics , mathematics , quantum mechanics , combinatorics
Purpose : To develop a comprehensive GPU‐accelerated, superposition/convolution based dose calculation engine supporting both a modern dual‐source MV model and an analytical kV source model Method and Materials : An analytical kV source model was developed for use with the Small Animal Research Radiation Project (SARRP), which analytically calculated the fluence to every voxel. The MV source model contains a traditional primary and an enhanced extra‐focal algorithm that efficiently accounts for leaf‐height effects. The spectral and attenuation effects of static beam modifiers were integrated into each source's spectral function. The total energy released per unit mass (TERMA) computation used back‐projection with optional exact multi‐spectral attenuation. Superposition/convolution was implemented using the inverse cumulative‐cumulative kernel and exact radiological path ray‐tracing with optional kernel tilting. Two superposition variants were implemented and benchmarked. Multi‐resolution superposition approximates true, solid angle ray‐tracing. Arc superposition increases the relative temporal TERMA sampling. Results : The kV source model performance was integrated into the TERMA computation, increasing performance in a volume resolution and collimator size dependent manner. MV source model performance was <9ms (data dependent) for a high resolution (400 2 ) field using an NVIDIA GeForce GTX 280. Our TERMA GPU implementation was improved, doubling performance. Our GPU Superposition implementation was improved by ∼18% to 0.058 s and 1.186 s for 64 3 and 128 3 water phantoms; a speed‐up of 80–142x over the highly optimized Pinnacle 3 (Philips‐Madison, WI) implementation. Pinnacle 3 times were 8.3 s and 94 s, respectively, on an AMD Opteron 254 (2 cores, 2.8GHz). Extensive accuracy benchmarks indicated that Arc Superposition increases dose computation efficiency and that Multi‐Resolution Superposition slightly under estimates MV intra‐patient scatter. Conclusions : We have completed a comprehensive, GPU‐accelerated dose engine in order to provide a substantial performance gain over CPU based implementations.