AIRWC – Accelerated Image Registration with CUDA

This is a project to develop very fast 3D non-linear Medical Image Registration code using GPUs.

The present code uses the NVIDIA CUDA programming language and runs on PC’s equipped with suitable NVIDIA GPU’s.

The current version is AIRWC 5.0 and supports 12 parameter affine transformation and BSpline based non-linear registration with grids of up to 131x131x131 control points (experimentally up to 259x259x259 for SumSQ only).

For typical MRI images of dimensions 240x256x176 affine registration takes a few seconds and full B-Spine registration about 5 minutes using a GTX 285 GPU.

Please e-mail me, Richard Ansorge, ( rea1@cam.ac.uk ) if you are interested in a copy of the code. The code would be made available for non-commercial use only and on an “as is” basis and without warrantee of any kind. Click Here to Get AIRWC (source5.0 is the most recent version)

At present the available cost functions are Normalised Correlation, suitable for intra modality imaging and Normalised Mutual Information and Sum of Squares.

A conference report is here: IEEE MIC 2009 Conference Report