Combining medical images from more than one modality - MRI and CT scans, for example - can enhance the quality of the final image. Unfortunately, generating a computer-enhanced image from multiple sources is far from straightforward.

Accurate image registration - the alignment in 3D space of images obtained at different dates or using different imaging devices - is key. Once the images are correctly aligned over one another, a radiologist can more easily detect structural changes such as the growth or shrinkage of tumours.

"This alignment of images both improves the accuracy of interpretation and improves radiologist efficiency, particularly for diseases like cancer," says Mayo Clinic radiology researcher Bradley Erickson. But when three dimensions and millions of pixels are involved, the task becomes increasingly complex - and higher processing speeds become essential.

To address this issue, the collaboration ported Mayo Clinic's image registration application onto IBM's BladeCenter QS20 - a high-performance blade based on its multicore Cell Broadband Engine (BE) processor. The Cell BE has nine processor cores on one chip connected with a high-bandwidth element interface bus, which IBM says enables "supercomputer-like" performance.

For this project, the researchers used the BladeCenter QS20 to run the registration application on 98 sets of images. The team adapted a "mutual-information-based" 3D linear registration algorithm application optimized for Cell BE and completed the registration for all sets of images in 516 seconds. No registration took more than 20 seconds.

In comparison, running the image registration application on a typical processor configuration required approximately seven hours to complete the registration of all 98 sets of images.

The 3D linear algorithm finds the best spatial positioning to maximize the amount of information gathered from the two images, thereby optimizing sampling quality while reducing sampling time. Greater efficiencies were achieved by caching data in cuboids so that the image sampling did not waste pixels. When the sampling ratio was comparatively low, the team packed the sampled moving pixel images in a contiguous fashion (an image stripe) to speed retrieval when needed.

"This is all about taking technology innovation, collaborating with our customers and applying it to help them directly benefit their patients," said IBM's Shahrokh Daijavad. "This improvement with the application running on Cell will allow Mayo's doctors and radiologists to achieve in seconds what used to take hours, which in turn will significantly decrease the wait time and anxiety for a patient waiting on news from the doctor."

Mayo Clinic and IBM presented the results of this study at the IEEE International Symposium on Biomedical Imaging in Washington, DC, last week.