Novel GPU-Based Pencil Beam Dose Calculation Including Beam Characteristics
A Sullivan1*, (1)SU-E-T-2 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: A fast, accurate pencil beam dose calculation method for IMPT can be used to produce higher quality treatment by enabling many different treatment plans to be tested and compared rapidly. It is important that the dose calculation correctly include the effects of the finite SAD, and changes in beam emittance associated with changes in energy and treatment port.
Methods: We exploit the combined capabilities of both CUDA and OpenGL to propagate a set of Gaussian pencil beams decomposed into weighted superposition of elementary Gaussian beamlets. In alternating phases the beamlets are advanced using CUDA to compute the beam propagation and OpenGL to compute the energy deposition into voxels. Using tabulated integrated Bragg curves and beam divergences obtained from measured machine-specific data, CUDA is used to compute energy lost by the beamlets due to the stopping power of the tissue obtained by conversion from CT, increases in beamlet size using the differential Highland formula to model scattering, and changes in the beamlet shapes due to finite SAD and beam emittance. The OpenGL geometry processor uses the beamlet centroid position, size, eccentricity and orientation to generate oriented rectangular geometry that is projected onto dose volume slices. The OpenGL fragment processor uses a scaled Error Function texture map to deposit beamlet energy into voxels. Both protons and carbon ions are easily handled as is the contribution of large angle scattered ions.
Results: Using 25 beamlets per pencil beam, the dose can be computed a square grid of 24x24x36 beam spots in a 90x90x250 mm3 voxel volume (1.0 mm voxel size) in 4.1 seconds using nVidia GTX580 graphics card with high accuracy.
Conclusion: We have developed a novel algorithm that exploits the strengths of both CUDA and OpenGL to enable fast, accurate dose calculation from ions in heterogeneous, high resolution CT volumes.