Labelconvert and connectome visualisation


#1

Hello,

I’m having troubles with labelconvert and using the connectome tab from mrview.

1- In labelconvert, I would need some help regarding the parameter ‘lut_out’.
From what I understand, however it’s called ‘out’, it’s an input that should contain some kind of template to guide the modifications that I want to have in my parcellation.
Is that correct ? So I have to manually modify my indices to increment from 1 everytime ?
I have tried the format that you suggest in ‘labelconvert: Explanation & demonstration’ with the colors however labelconvert doesn’t recognize the format.

2- Regarding the visualisation of my connectome:
When I load my parcellation (image output of labelconvert) in Node image, the nodes that I get are all aligned on one sagittal plane.
Am I missing any extra step ? For now in labelconvert lut_out I use the same lut as in lut_in (index and name) but with an increment from one in the indices.

Thanks for your help!

Céline


#2

Hi Céline,

  1. Yes, that argument is not actually an “output”; I should think of a different name for it. It’s the lookup table to which you want your output image to conform.

    Generally if you’ve taken a lookup table that MRtrix3 can already read, and simply modified the indices, then there shouldn’t be a problem. However it only takes one slightly erroneous line for the code to detect an inconsistency in the formatting and hence be unable to read the file: It’s looking for a consistent number of columns in each row. While the list of supported formats isn’t in the documentation, you can see the possibilities in the code here. I do have code somewhere for making errors in parsing here more verbose in response to this thread, but I can’t recall where it went; if you get stuck, you can try posting the contents of the file here.

  2. It took me a moment, but I think I know what’s going wrong here. The connectome tool determines the centre of mass of each node in 3D, and draws the node (by default a sphere) at that location. For all of the nodes to be aligned on a sagittal plane would require that the centre of mass of every single node lies on a sagittal plane. There are only two ways in which this could occur:

    • (unlikely) Every node only contains voxels on a single sagittal plane.

    • (likely) Every node is labelled identically in both hemispheres, resulting in a centre of mass that lies on the mid-sagittal plane.

    The latter I believe I’ve observed in FSL atlases, and was the reason why I haven’t provided lookup tables for those atlases within MRtrix3: That labelling can’t be resolved for use with tck2connectome using labelconvert, explicit image manipulation is instead required. I imagine that you would need to find / derive a hemisphere mask image, and add some integer number to node values within that masked hemipshere, in order to get a unique integer for each node (including homologous regions between hemispheres).

Rob


#3

Hi Rob,
thank you very much for your help!
Yes, you are right my brain is identically labelled in both hemispheres !
Céline