Ordering Regions

[SandyaS72]

Is there an accepted ordering of brain regions that puts regions that are similar near each other? In other words, if you went down the ordering, you'd be progressing down a "spectrum" of sorts?

[indigorose1]

The problem with that is the brain regions share a 3D relationship with each other. Even if you just focus on the cortex and lay it out flat, it's still 2D—so how do you systematically number the regions so that they're close to their sister regions in both the x and y dimension? You could do a hilbert curve or a z-order curve (tbt Will & Greg's connectomics class), but the brain regions aren't divided up into nice squares so the curve would be distorted. I'd be interested to see a computer simulation that optimised the structure of the curve so that regions that are physically close to each other on the cortex have the best representation of their closeness in their numbering.

[indigorose1]

But then if you consider sub-cortical structures all hell breaks loose because who's to say that V1 is closer to or farther from the thalamus than S1 if they're non-contiguous. You'd only have 3D space as your metric, which probably isn't a very good metric in this case.

[wrgr]

+1 Yea, not really a standard.

We sometimes look at really coarse things like hemishere (intra- v inter-hemispheric), which is something. There are a lot of different atlases and options, and you can reorder as you prefer with the metric you prefer for your task. Someone who wants a coarse functional parcellation is going to have a very different idea of closeness than someone doing detailed structural analysis...

It's also a problem in EM data...how do you pick a point representation of a neuron that spans a large region? Somata centroid?

[jovo]

we are starting to play with matrix orderings, which can be fairly powerful, efficient, and useful. for example, choosing the ordering to make the adjacency matrix as "banded" as possible.

On Wednesday, February 4, 2015, William Gray notifications@github.com wrote:

+1 Yea, not really a standard.

We sometimes look at really coarse things like hemishere (intra- v inter-hemispheric), which is something. There are a lot of different atlases and options, and you can reorder as you prefer with the metric you prefer for your task. Someone who wants a coarse functional parcellation is going to have a very different idea of closeness than someone doing detailed structural analysis...

It's also a problem in EM data...how do you pick a point representation of a neuron that spans a large region? Somata centroid?

— Reply to this email directly or view it on GitHub https://github.com/Statistical-Connectomics-Sp15/intro/issues/47#issuecomment-72887017 .

the glass is all full: half water, half air. openconnecto.me, we're hiring! https://docs.google.com/document/d/14SApYAzxF0Ddqg2ZCEwjmz3ht2TDhDmxyZI2ZP82_0U/edit?usp=sharing , jovo.me, my calendar https://www.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

[SandyaS72]

Actually that's why I was wondering- if there was a way to shuffle the rows and columns of an adjacency matrix so that the ordering "made sense" and anywhere in the matrix, rows next to each other represented "similar" regions..... though I guess you could go the other way and use the matrix to figure out the ordering too

[jovo]

http://www.mathworks.com/help/matlab/math/sparse-matrix-operations.html?searchHighlight=banded%20matrix%20ordering#f6-10934 lists a bunch of sparse matrix orderings, note that sparse matrices naturally represent adjancency matrices of graphs. probably will be fun and informative to play around with some of these functions.