Usage tips

  • To switch between the individual subject and the template, use the space bar to toggle back and forth. The text field above the image also enable this (left and right clicks), but it also shows more images, like distortion fields and basis functions.
  • When the affine alignment page is shown, the tab key toggles between adjustment of parameter values versus adjustment of value increments. You can increase or decrease a parameter increment, and then adjust the parameter manually as desired.
  • Even though mutual information is designed to enable registration across contrasts and/or modalities, it is always better to use similar contrasts when possible (e.g., T2-weighted for both subject and template). If you are constantly registering across contrasts, consider replacing the template with the multi-subject average.
  • Suppose you want to align relatively distorted EPI data to anatomy, and you know that the EPI data will not have a lot of intrinsic contrast (i.e., no iron oxide agent in use, so you have low-contrast BOLD data). In this case, acquire a relatively quick undistorted anatomical scan (e.g., T1 or fast-spin-echo T2) using a matched set of slices, then use this anatomical volume for a first alignment step. Then, because the EPI slice-selection step is the same as for the anatomical, remove the z and size-z parameters from the affine adjustment set, and align the EPI data using the first alignment as the starting set.
  • When creating a template, try to use dimensions that are divisible by 8. While this isn't a requirement, it does allow an integral number of non-linear distortion basis function to fill the field of view using up to 3x down-sampling in resolution.
  • A disadvantage of the "roll-off" filter is that dividing an image by a low-pass version can produce a ring of high intensity near brain surfaces when the template has been previously stripped of intensity outside the brain. To prevent this, replace skull-stripped templates (when possible) by your own templates that do not have extracerebral tissue removed.
  • Sometimes it may be frustrating to manually initialize alignment if the volumes are large and/or in different orientations. You can down-sample images to really speed up the process. Also, make sure that the overlay filter has not been applied so that tissue outside the brain mask can be seen. This doesn't work, however, if the template already is zero outside the brain.  See bullet point above.
  • Which is more important in terms of CPU size: the size of the the source/individual volume or the target/template volume? The answer is that CPU depends more strongly on the size of the template volume. A loop is performed over all voxels in the template space for each iteration, and then the transformation identities a point in the source and interpolates nearest neighbors. So, don't use excessive padding (larger FOV) on templates, but do include just a bit of space around the brain to enable smoothing without wrap-around artifacts.
Joseph B. Mandeville, Athinoula A. Martinos Center for Biomedical Imaging at MGH/MIT/Harvard