Run the diffusion processing pipeline at low resolution

When first running diffusion MRI data from a new protocol through the pipeline, you will probably discover modifications that must be made in order to correct for orientation problems, registration problems, or incorrectly reported b-values. Therefore it's valuable to use a lower-resolution version of your data while troubleshooting the dataset, and only run at full (or upsampled) resolution once you've coded up automatic fixes for these issues.

The first step of the pipeline is to convert the DWIs from the original format of the data (in which it was given to us by a collaborator) to our standard format, MRIimage.

Once the DWIs are in MRIimage format, the next step is to use diffusionResize.py (currently available at /map/gfx0/common0/diffusion/Interface/bin/data) to upsample them and crop/pad them to a standard physical volume. We can also use this program, however, to downsample and crop out a smaller region of the brain. Here is a recommended command:

rm -rf dwi/resized
mkdir -p dwi/resized
diffusionResize.py dwi/original dwi/resized 210 210 210 80 80 80

This crops out a 210mm x 210mm x 210mm region in the center of the dataset and resamples it to fit an 80x80x80-voxel cube. This is sufficient size and resolution to recognize many distinct structures in the brain while using about 10% as many voxels as a typical run. Cropping the volume further is not recommended, as tubegen culls any streamtubes that escape from the sample volume.

Once you've performed this alternate second step, proceed with the rest of the pipeline, and be sure to use $G/bin/mriatlas to look at the output of each step!