Hello!
I’ve been using MRtrix for a while. Lately I’ve noticed that commands like dwi2response and dwi2fod spend an unexpectedly long time just getting the data into memory. Watching the process, nearly all the wall-clock time goes to this step:
mrconvert Diffusion/data.nii.gz -fslgrad Diffusion/bvecs Diffusion/bvals diffusion.mif -force -strides 0,0,0,1
My understanding is that the -strides 0,0,0,1 reorders the data so the volume axis is contiguous, making downstream per-voxel access faster. My data is HCP (~145 volumes, 1.25 mm iso, gzipped NIfTI), with data on local storage.
Is this conversion something I can skip or speed up? Ideally I’d like to avoid paying this cost repeatedly across many subjects. Thanks!