Could we use our own atlas for brain segmentation using FastSurfer?

[AXu0511]

Question/Support Request

Maybe I have missed it, but could we use our own atlas for brain segmentation using FastSurfer? If possible, could you please tell me specifically how to operate it? Looking forward to your reply soon.

[LeHenschel]

Hey,

overall, there are two potential ways you can do that.

1.) you can retrain the networks on data with your own atlas as ground truth. 2.) if the atlas is defined on FSAVERAGE you can align the generated surfaces to it with sphere.reg and map the labels to the subjects (similar to how the original surface parcellation is done in FreeSurfer e.g.).

Kind regards, Leonie

[AXu0511]

Hello,

If my atlas is defined on FSAVERAGE, can I use my atlas for segmentation in FreeSurfer? If so, what command should I use to replace the atlas? If my atlas is not defined on FSAVERAGE, is there any solution to transform it to be defined on FSAVERAGE? Kind regards, AXu

[LeHenschel]

Can you be a bit more specific what you actually want to do in the end (Do you want FastSurferCNN to return your segmentations or you are fine with running the surface pipeline as well?)?

If your atlas is on FSAVERAGE you can use the mris_ca_label from FreeSurfer to get a parcellation of the surfaces: [mris_ca_label](https://surfer.nmr.mgh.harvard.edu/fswiki/mris_ca_label) ../label/lh.cortex.label -aseg aseg.presurf.mgz <subjid> ?h ?h.sphere.reg $ATLAS_HOME/?h.YOURATLAS.gcs ../label/?h.aparc.YOURATLAS.annot

To prepare such an atlas from annotated labels see https://surfer.nmr.mgh.harvard.edu/fswiki/mris_ca_train

[AXu0511]

In fact, I hope to obtain a segmentation map that is consistent with the segmented regions of my atlas through FastSurfer or FreeSurfer. I noticed that using FastSurfer or FreeSurfer can generate many different segmentation maps, but I am not particularly interested in the structures of certain brain regions in these results. Therefore, I am curious if there is a way to use my own atlas as a segmentation template to obtain segmented maps with consistent segmentation regions.

[LeHenschel]

Ok, if you do not want to retrain you can use the commands I posted above to get your own atlas segmentations. This works with both FastSurfer and FreeSurfer output.

[AXu0511]

Thanks a lot. I will try both methods and If I successfully obtain the segmentation I want, I will close this issue.

[RDoerfel]

Hi, I'm sorry for hijacking this issue, but I think my issue is rather similar, and I would like to confirm.

Is it possible to get parcellations and respective stats for other atlases than the DKT atlas? More specifically, I'm talking about the HCP (Glasser) atlas.

[LeHenschel]

Yes, the commands above should work for any atlas in theory (as long as it is defined on FSAVERAGE, as the registration of the subject surfaces to it already exists (=sphere.reg)).

[RDoerfel]

Awesome, thank you!