Florian Knoll1, Manuel Freiberger1, Kristian Bredies2, Rudolf Stollberger1
1Institute of Medical
Iterative image reconstruction of undersampled data from multiple coils has shown a huge potential for a wide range of applications during the last years. One of the main restrictions of these methods is the prolonged image reconstruction time. Parallelized implementations on graphics hardware were recently discovered as a feaseable method to significantly speedup image reconstruction. While programming graphics hardware was simplified significantly with the introduction of dedicated libraries for general purpose computing like CUDA or OpenCL, efficient implementation, especially concerning memory management, is still a challenging task. The goal of this work is to introduce an open source library designed for image reconstruction on GPUs. It is based on highly efficient implementations of numerical methods, but also includes code for iterative MR image reconstruction as well as a framework for finite element calculations and applications for Fluorescence Tomography. The results from this work illustrate the pronounced computational speedup with the GPU implementation. As the toolbox is an open-source project, all algorithms can be used either as a black box for the reconstruction of new data, or as a basis for own implementations for similar problems. The latter case is facilitated by the object-oriented and templated design which provides a well defined structure for extensions and allows for maximum code reusability. The intention behind the release of this open-source library is to alleviate the usage of the huge potential of graphics hardware in medical image reconstruction.