Image dimension reduction in population_template

Hey there,

I’ve been trying to create a longitudinal FA template across two time points using population_template.

population_template <output dir/fa_template.mif> -mask_dir -tempdir -nthreads 4 -type rigid -voxel_size 1.3

This code produces an FA template without problems. However, the image dimensions of this FA template is reduced in comparison to the FA input images. Both FA input images have a dimension of 168x168x95 at an isotropic voxel size of 1.3. In comparison, the FA template image has the dimension of 124x164x116.

I can’t figure out where this mismatch comes from, at which point in the population_template script the dimensions become reduced, why and how I could avoid it. Can anyone help?

Thanks so much in advance,
Caroline

The voxel grid in population template is refined in multiple steps, with your parameters it starts as the voxel grid that contains all scanner locations in all input images*. It is refined to only span the union of the dilated masks of the input images and then to cover that of the dilated masks translated to a common centre of mass. This is to keep file sizes as small as possible.

Guessing the motivation behind your question, note that the template image grid will most likely not correspond to the input image grid (unless there is no difference between the input images) and even if it did, the information in the same voxel coordinates would not necessarily correspond to each other. To compare voxel-values between template and input data you’d need to warp either to the other’s space or warp both to a common space (for instance using mregister and mrtransform as described in the FBA pipeline).

*For input images that have the same number of voxels in each dimension but different transformation (mrinfo) you’d end up with a larger template voxel grid to cover all scanner locations in both images.