dMRI processing

[tsalo]

1. QSIPrep
2. QSIRecon
   • Enable microstructure maps

Open questions:

• What resolution should I use for the QSIRecon derivatives? We want maps to be comparable across measures, and the highest-resolution measures are 1 mm isotropic. Should I write out images in that resolution?

[tsalo]

@mattcieslak would you mind weighing in on the resolution question?

[mattcieslak]

Upsampling to 1mm will cause qsiprep to use linear interpolation to prevent artifacts, but we will probably lose some crispness too. I usually try to keep the resolution of the outputs from qsiprep as close to the inputs as possible

[tsalo]

That makes sense, but I can't compare myelin measures on a voxel-wise basis if they're not at the same resolution. The alternative would be to downsample the higher-resolution images to the lowest native resolution across modalities, which would be 2 mm isotropic from the ihMT scans.

[mattcieslak]

You may end up wanting to try it both ways. There aren't too many subjects here, right?

[tsalo]

We'll ultimately have 20 maps per derivative: one for each of 10 subjects from each of the two sessions. The goal is to have a detailed analysis plan before we start analyzing the data, so if we do it both ways we'll have to report on both.

I can hold off on making this decision until the NiBS folks have a chance to meet and discuss the analysis plan though. Ted might not even want to do a voxel-level analysis.

[tsalo]

Currently the only things we haven't figured out are (1) resampling to do for the analyses, which depends on the analyses we end up doing, and (2) how to calculate the g-ratio, which I'm going to look into.