Necessary adaptations to processing for ex-vivo / animal data

Hello! I have a rather general question to the DWI experts here on which steps to consider when applying mrtrix to high-resolution ex-vivo (single or multi-shell) animal data.

I read Problem with dwi2response msmt_5tt (amp2response) command for ex-vivo monkey data, Estimate response function for multi-shell ex-vivo marmoset data and Multi-tissue CSD and it seems that the mrtrix steps of response function estimation and FOD fitting can be used in a rather straight-forward way on non-human data as well. In general terms, is that also the case for tractography and track density estimation (eg. SIFT2)?

I would like to quantitatively compare the connectomes of different animals and humans (data very similar or identical in quality to the links above) - right now I’m generating 5tt segmentations semi-manually for each animal (I merge sub-cortical and cortical gray matter as per this comment), use 3-tissue (dhollander (2019)/msmt_csd on multishell or MRtrix3Tissue dhollander/ss3t_csd_beta1 for single shell data) CSD for FOD estimating, run tractography using dynamic seeding and compute sift2 weights for the resulting tracts (20M). I then compute a sift2-weighted TDIs for quality control and extract tracts using masks of subcortical structures (thalamic nulcei etc.) as selection masks in tckedit.

Now, as I’m by no means an expert in MR/diffusion phyics, I was wondering if and to what degree this processing is valid in ex-vivo data of different species, and what would need to be done to make the processing more quantitative. I know this is a very broad topic…

Hi Ernst,

It’s tough to decide the appropriate scope at which to address this question. I’ll see how I go, but let me know if what you’re actually looking for is +/- of this.

I’ll take from your description that you understand / “trust” the operation of CSD / the AFD metric, streamlines tractography in its most fundamental form, and the model underlying SIFT(2). From there, I’m going to flip the idea on its head: what would need to be different between humans / animals, and/or in vivo vs. ex vivo, in order for the operation of these methods / metrics to be invalidated?

  • I’m going to skip over the resolution of fibre orientations via CSD, and basic reconstruction of trajectories via streamlines tractography, since it’s difficult to envisage such becoming fundamentally wrong under one of these changes.

  • Does the magnitude of the (white matter) FOD increase if the underlying axon density increases? Historically this would be a question of whether the DWI signal intensity increases with greater axon damage; it’s more complicated now with how multi-tissue CSD operates, but in general I think this should be true.

  • For SIFT(2), the model is predicated on AFD (covered above) and tractography being appropriate. I’ll also throw in here that axons don’t synapse / terminate in the white matter, which is more specifically an ACT thing but has ramifications for SIFT(2); this can be broken in instances of pathology, but I’m working on that. Finally, there’s an assumption that axon diameters (not of individual axons, but bundle-wide statistics of such) don’t vary wildly along the lengths of bundles; this is difficult to violate from a physical perspective since diameters can only increase so much within a constrained volume.

So from a preliminary consideration I don’t see any major hurdles that mean that these specific methods would not be applicable whereas those methods on which they are based are. ex vivo can have different behaviour to in vivo with respect to tissue decomposition / response function estimation, but those are upstream of the question at hand. While I’ve not worked with a lot of animal data, you may want to consider how to deal with “cortical” vs. “sub-cortical” grey matter in ACT, since the ratio between white matter bundle width and cortical depth can be very different compared to humans and therefore permitting streamlines to project into the grey matter may provide additional information in that case.

That’s all for now
Rob