high performance computing on graphics processing units: hgpu.org

The Edinburgh Cancer Centre at the Western General Hospital in Edinburgh is doing research on image analysis for predicting lung fibrosis induced by radiation as part of a treatment plan. They are developing a MATLAB code to analyse three dimensional Computed tomography (CT) images of patients but, because a standard three dimensional CT image is a large data set to be processed, the original MATLAB code runs very slowly and takes a long time to produce a result. This project tackles the challenge of processing large data sets of three dimensional CT images and accelerating the original MATLAB code. The project focuses on the most computational demanding part of the code, which is filtering the three dimensional CT images with a Gabor filter. We improve the image filtering algorithm with a Fast Fourier Transform, and apply multi-cores and GPU techniques for parallelisation. In this project, three GPU technologies are used to optimise the original MATLAB code. The first one is the MATLAB Parallel Computing Toolbox. The second is AccelerEyes’ Jacket, which is a GPU based accelerator for MATLAB. The third is CUDA from the NVIDIA Company, being hybrid programming with MATLAB through MEX files, which are MATLAB Executable files. Combining MATLAB with GPU techniques and algorithm improvements, the original MATLAB code is optimised largely, and has achieved a speedup of 120.