3D labelled atlas rather than 4D probablistic?

[rtybanana]

Hi,

As I understand it, the DiFuMo atlas is currently accessible as a 4D probalistic atlas with each region on a different volume and voxel values corresponding to the probability that it belongs to the ROI marked by each volume. For my purposes, I need the atlas to be in a 3D format where voxels are labelled with a number corresponding to the region ID; however, I've done some research but can't find any documented way of converting between the two.

Is there a way that you know of to convert from a 4D probablistic to a 3D labelled atlas, perhaps using FSL?

Many thanks, Rory

[bthirion]

A 3D atlas would be convenient if a voxel belongs to one and only one component, which is not the case with DiFuMo (Let's call this a parcellation). You can do an argmax of the atlas values per voxel, but this loses many of the benfits of using DiFuMo. It would make more sense to use e.g. Schaefer atlas in that case. HTH,

[rtybanana]

Thanks for your quick reply.

So conceptually there is nothing wrong with argmax-ing along the 4th dimension of the provided atlas as long as we don't care about losing the overlapping ROI information in this case?

Essentially, what we are trying to do is combine functional and structural connectivity into a single structure for analysis. We're trying out a bunch of atlases in this research because some are more suitable for functional and others more for structural. We have a working pipeline for extracting timeseries from our functional scans using DiFuMo, but got caught up with doing the same for our structural scans because the atlas was expected in a 3D format.

[bthirion]

On 03/04/2021 00:57, Rory Pinkney wrote:

Thanks for your quick reply.

So conceptually there is nothing wrong with argmax-ing along the 4th dimension of the provided atlas as long as we don't care about losing the overlapping ROI information in this case?

Yes, I'd say so.

Essentially, what we are trying to do is combine functional and structural connectivity into a single structure for analysis. We're trying out a bunch of atlases in this research because some are more suitable for functional and others more for structural. We have a working pipeline for extracting timeseries from our functional scans using DiFuMo, but got caught up with doing the same for our structural scans because the atlas was expected in a 3D format.

OK. I hope that you obtain meaningful results from this !

Best,

Bertrand