AGILE: GPU Image Reconstruction Library

AGILE (Environment for Linear and non-linear Image reconstruction using GPU Acceleration) is an open source library for GPU accelerated reconstruction problems in medical imaging. It originally included reconstruction code for both MRI (Figure 1) and Fluorescence Tomography, as well as general purpose linear algebra routines. The current version that is made available for download here only includes the MR reconstruction part to facilitate easier installation. If you are interested in the complete version of the library, just contact us. (We have also made a MATLAB version available for download.)

Figure 1: Reconstruction of undersampled radial data from 48, and 32 radial spokes (256x256 matrix). Conventional regridding reconstruction (Left column) and AGILE iterative TGV image reconstruction (right column). The corresponding reconstruction times on an NVIDIA GTX 480 were 8.76s (48 spokes) and 6.36s (32 spokes).

Contributors (sorted alphabetically):

• Kristian Bredies (University of Graz)
• Gerald Buchgraber (Graz University of Technology, now at Datenkraft IT-Consulting)
• Manuel Freiberger (Graz University of Technology, now at Anton Paar)
• Andreas Huber (Graz University of Technology)
• Florian Knoll

Version History:

1.2 (14.03.2013): Extracted MR reconstruction part. Changed support to CUDA 5.0 (CUDA helper functions instead of SDK functions)
1.11 (13.04.2012): Minor changes in documentation and readme.
1.1 (16.01.2012): Included CPU reference implementation for TGV.
1.0 (28.07.2011): First release of AGILE.

Related Publication:

If you use this library or the provided data in your publications, please cite:

Freiberger, M.; Knoll, F.; Bredies, K.; Scharfetter, H.; Stollberger, R.: The AGILE library for image reconstruction in biomedical sciences using graphics card hardware acceleration: Computing in science & engineering 15:34-44 (2013).

For any questions, comments or contributions, please contact Florian Knoll (florian.knoll@nyumc.org).