Evaluating the Performance of Processing Medical Volume Data on Graphics Hardware

Abstract. With the broad availability and increasing performance of commodity graphics processors (GPU), non-graphical applications have become an active field of research. However, leveraging the performance for advanced applications combining hardware and software implemen-tations is more than just efficient shader programming: the data transfer is often the main limiting factor. Therefore, we will investigate in the applicability of commodity graphics hardware for medical data process-ing, and propose a GPU-based framework for representing computations on volume data. Also, we will show the clear performance gain of differ-ent operations compared to CPU algorithms and discuss their context. Not only the much higher performance of hardware implementations is attractive, but also the fact that the computation results can be visual-ized directly, i.e., without introducing an overhead and thus allowing for literally interactive applications.