GPU Beamlet Dose Calculation

Ideated and implemented a novel approach to parallelizing beamlet dose calculation at both the intra- and inter-beamlet levels. Resulted in >24x speedup in calculation time over an existing GPU CCCS dose calculation method that only parallelized at the intra-beamlet level with <2% average error and <5% max error to single-beamlet dose.

Involved custom CUDA implementation of dose calculation that exploits assumptions of lateral compactness of dose scatter in MeV X-Ray beams. 3D geometric transformations of original CT image are used to place required information in more GPU-compatible form for efficient convolution.