The Rise of Image Registration

In a typical clinical trial that makes use of imaging, a subject is often scanned multiple times over a period of weeks or months. In-order to get a precise assessment of how well the drug being tested works, you need to detect subtle changes in the patient which might have been caused by the drug. While radiologists are very skilled at detecting abnormalities in images, they are not so good at measuring change precisely. This is especially true if the changes are small, as, for example, in Alzheimer's disease where the brain shrinks by a tiny fraction each year.

The first step in comparing two images is to bring them into spatial alignment, a process referred to as registration. One of the original ways of aligning the images is to simply introduce an external marker attached to the subject such that it is easily detectable in the images. Since the early 1990s, however, computer algorithms that can register the images without using markers have been developed and these are now widely used.

The founders of IXICO, Derek Hill, Jo Hajnal, David Hawkes and Daniel Rueckert, pioneered several image registration techniques in the 1990s, and demonstrated the value of image registration in diagnosis, planning treatment and monitoring disease progression.

Here’s a few words from IXICO’s CEO, Derek Hill:

“Image registration is now such an important sub-field in medical imaging that there are now conferences dedicated to it. Two important branches of image registration are simple alignment of the images to correct for position differences (often called "rigid body") and image registration that tries to exactly map one image onto another by warping (often called "non-rigid"). In rigid-body registration, if you subtract the images after the registration, you can see areas where there has been change that may be due to disease progression or drug treatment. After non-rigid registration, however, the two images look identical and it is important to analyse the valuable information in the warps to get an indication of what has changed.

Image registration is not an end in itself but a means of making more effective use of medical images, especially where precise measurements are needed. Image registration algorithms are now commonly used in many clinical trial areas as a key step in the analysis, such as measuring brain shrinkage (atrophy) and the amount of blood flowing to cancer tumours.

However, image registration can also play a role in the trials well before making measurements, by ensuring the image data is of high quality. IXICO uses image registration algorithms to check that a subject was scanned correctly (in the right place in the scanner, and imaging the correct parts of the body), to ensure that the images with the same subject label are really of the same subject (accidental mislabelling of images is, regrettably, a common feature of clinical trials), and to detect whether scanners have gone out of calibration over time. In these situations, using registration methods on the data as soon as possible after it has been collected can enable potential bad data to be detected and dealt with before it is too late.”