high performance computing on graphics processing units: hgpu.org

Radiotherapy is one of the main cancer treatments used today. It is a complex process that relies on finding the cancer in the images of the patients with the most accuracy possible in order to minimize the radiation that the surrounding organs receive. Given that a typical radiotherapy treatment process lasts for 6 weeks, ideally, a system that performs this analysis in real time reliably would enable a better treatment process. The Western General Hospital and the Edinburgh Cancer Research Center have a prototype system based on texture features analysis of small lung and rectum cancer patients’ images that is currently being developed and used for their research. It was implemented in MATLAB and we used and compared four different technologies to optimize the software: the MATLAB Parallel Toolbox using a small computer cluster, the GPU computing support given by that toolbox, the Jacket add on for GPU computations for MATLAB and finally C/CUDA native function calls from MATLAB, all these GPU technologies were carried on using a NVIDIA Tesla C2050 card.