next up previous contents
Next: Outputting the Results Up: The Simple Coreg Tool Previous: Manual Coregistration   Contents

Automatic coregistration

Tina can perform the coregistration automatically, by optimising a metric that measures the similarity between the two image volumes. Two metrics are available: a chi-squared metric based on the alignment between edges in the images, and the Mutual Information measure (see Advanced Coreg Tool below). Both methods use simplex optimisation, a local optimiser. Therefore, a rough alignment must be achieved before automatic coregistration, to prevent the optimisation becoming trapped in a local minimum. This rough alignment can be achieved either by hand, using the manual methods specified above, or by loading in an AIR file specifying the rough alignment. If such an AIR file exists, enter its path in the AIR file field, and press ``input" in the simple coreg tool. The values specified in the AIR file will be loaded into the AIR parameters dialog box, and the result of applying the transformation to the source image volume will be displayed in the three coreg Tvs.

Prior to coregistration, select the degrees of freedom that the transformation model will possess using the ``Model" check buttons. Any combination of translation, rotation and scaling can be selected.

During coregistration an interpolation algorithm is used to resample the source image volume on the voxel grid of the target image volume. The interpolation routine to be used is selected via. the ``reslice" choice buttons. The routines available are

The chosen interpolation algorithm is also used to produce the displays in the three coreg Tvs, to produce the stored projections through the target image volume when the data is latched, and to reslice the data after coregistration (see below). In general, tri-linear interpolation provides a good balance between speed and accuracy. Sinc 5 interpolation will produce a more accurate result, but will require several minutes to perform an automatic coregistration. Note that the interpolation method can be freely changed at any time: for instance, the coregistration can be performed using tri-linear interpolation for speed, then the data can be resliced using sinc5 interpolation for increased accuracy.

In order to perform the automatic coregistration, press the ``Coreg" button. The edge-based similarity metric will then be optimised, the resulting transformation model parameters displayed in the AIR parameters dialog box, and applied to the source image in the three coreg Tvs. The values of the similarity metric before and after optimisation, and the number of iterations performed, will be displayed in the text window of the top-level tinaTool. Due to the nature of simplex optimisation, more accurate results (subject to a law of diminishing returns) can be achieved by re-running the automatic coregistration (by pressing ``Coreg" multiple times: little increase in accuracy will be obtained after the third run). The calculation of the similarity metric is dependent on the ``border" and ``blur" fields, described in the AIR parameters dialog box quick reference below. In general, a border of 5 is sufficient if a rough alignemnt has been achieved before coregistration. A ``blur" scale of 0.75 should be used for images of the same scene (same subject for medical image volumes). Note that, although coregistrations can be performed between different scenes (different subjects) the statistical interpretation of the similarity metric in this case is questionable, regardless of the similarity metric used. The blur scale should be set to a higher value (typically 3) in such coregistrations to accommodate the differences between the data (the source image volume) and the model (the target image volume).

The edge-based similarity metric may have problems coregistering image volumes in which equivalent edges have opposite signs. This situation may occur when coregistering MR and CT images. In this situation, use the ``Modality" choice in the advanced coreg tool (see below).


next up previous contents
Next: Outputting the Results Up: The Simple Coreg Tool Previous: Manual Coregistration   Contents
root 2017-11-20