Automatic image registration — the task of aligning or mapping multiple images into a common co-ordinate space — is an important investigative and diagnostic tool in medicine. It is a crucial operation for image guided surgery, and aids surgical planning, radiotherapy treatment planning, anatomical mapping of physiology and function, and assessment of disease progression and response to therapy.

To date, the enormous computational requirements, especially for non-linear 3D registration, have often precluded image registration from use in medical research and clinical practice. We present a new approach that leverages recent advances in the power, programmability, and data capacity of commodity graphics processing units (GPUs) to greatly accelerate 3D image registration, reducing execution to a clinically compatible time frame while simultaneously providing real-time visual feedback on the process.