Cortical + subcortical map

[daamartins]

Dear community

I am new to brainsmash. I am in need of generating surrogates from the DK parcellation for an imaging transcriptomics analysis but i am unsure about how to implement this from the examples. My idea was to create a distance matrix based on euclidean distances between the center of each pair of parcels in the atlas (including both cortical and subcortical regions). I am unsure about a few things: 1. Can one produce surrogate maps including cortical and subcortical parcels in the same vector/distance matrix? 2. What about the parcels in the middle wall of each hemisphere (which will have a close distance but actually belong to different hemispheres); is this an issue? 3. What would be the most appropriate way of generating surrogate maps that include both cortical and subcortical regions from both hemispheres? Did anyone face this experience before? Thanks in advance for the help/advice

[jbburt]

Welcome!

In regards to each of your questions:

1) Yes, it is possible to do this, provided you have a parcellated brain map and a pairwise parcel distance matrix. To my knowledge, whole-brain surrogate maps have only been generated at the voxel level using BrainSMASH, but it should be just as easy (in fact, easier) to do at the parcel level.

2) The medial wall is indeed problematic (as we described in our paper). It would probably be best to exclude those parcels since their distances to other parcels will be very unreliable. It is recommended to use geodesic distances in cortex for this reason -- namely, that distant areas in the anatomical brain space will appear proximal in 3D Euclidean space due to the geometry of the cortex.

3) I think your first option is to simply do what you have described (ie, create a whole-brain Euclidean distance matrix and proceed). This is obviously problematic as it doesn't account for bilateral symmetry or the convoluted geometry of the cortical surface, but it should give you a more reasonable null than uncorrelated white noise. The second approach you might try (and I'm just throwing out ideas here -- I don't think this has been done) is generate one unilateral subcortical surrogate map (using a Euclidean distance matrix) and one unilateral cortical surrogate map (using a geodesic distance matrix), combine the two de-meaned maps (across cortex+subcrotex, to get a unilateral whole-brain map), and reflect them contralaterally (to get a bilateral whole-brain map). The assumptions underlying this approach would be that spatial autocorrelation is significantly greater within relative to across gross anatomical structures (which is a reasonable assumption, I think) and that brain maps are bilaterally symmetric. You could avoid this bilateral symmetry assumption by averaging your empirical brain map across hemispheres before comparing it to surrogate maps.

I hope this helps. For one other suggestion you might try looking at this issue opened by Ross Markello, where he suggests setting some elements of the distance matrix to infinity.

Josh

[jbburt]

One more thought: if you constructed surrogate maps for cortex and subcortex (independently), and then used the resampling feature of BrainSMASH to resample surrogate map values from your empirical brain map, this would yield surrogate maps which preserve the structure-specific distributions of values (in other words, the cortical piece of your surrogate map would contain the same values as the cortical piece of your empirical map, and same for subcortex). Doing this would ameliorate any complex issues arising due to differences in, for example, mean values or variance across these two subdivisions of the brain.