Image Registration

Introduction

The Image Registration tool is used to bring images into alignment, for example:

• When multiple scans are performed on the same subject, but the the image planes are not in the same locations on every scan (intra-subject registration). The multiple scans may be performed at different times on the same medical scanning device, or they may be collected using a different modality (e.g. MRI, CT, SPECT, PET).
• When you have scans of the same type performed on different subjects (inter-subject registration). You may want to register one of your scans to an atlas, or register several scans to create an atlas.

You can choose the type and number of degrees of freedom to meet the demands of your problem. The degrees of freedom determine the different types of image distortion allowed to obtain a good match.

The Image Registration tool takes one image as the "base" image (the image where the scan planes are in the desired location). The scan planes in one or more other images are then re-aligned so that the scan planes in the registered images lie in the same anatomical locations as in the base image.

Depending of the selection of , the registered image can either be:

• automatically loaded as an overlay image in Jim.
• saved to disk, and then loaded into Jim or analysed further as desired.

The Image Registration tool works by calculating a "cost function" that reaches a maximum or minimum as images are brought into alignment. The alignment is adjusted iteratively until the best value of the cost function is found.

The deformations allowed when bringing the two images into alignment can be simple (such as a translation and a rotation), more complex (such as allowing scaling and shears), or they can be very complex with hundreds or thousands of degrees of freedom with non-linear registration.

There is also a version of the Registration tool that is specifically "tuned" for time-series registration where the image to be registered are from a single subject, and where the images are already in a fairly close alignment. This situation is often encountered in functional imaging, where a series of images is collected an there is very little patient motion between the images. Use of the Time Series Registration tool allows much quicker registration of such data.

To start the Image Registration tool, select it from the Process menu:

This brings up the Image Registration tool:

You now need to decide whether you are going to do affine or diffeomorphic (non-linear) registration by selecting the appropriate tab:

Affine registration:

• Allows simple transformation ranging from a translation up to rotations, scaling and shears with a maximum of 15 degrees of freedom.
• Is suitable for intra-subject registration where there are no major changes in anatomy, and simple inter-subject or atlas registration.
• Is relatively fast.
• Is fairly robust.

Diffeomorphic (non-linear) registration:

• Allows much more complex deformations.
• Is suitable for forcing conformation of anatomy between subjects or with a template image.
• Needs careful setup and oversight, since there may be hundreds or thousands of degrees-of-freedom, with much more scope for the registration to fail.
• Is computationally intensive.