Freesurfer segmentation

[parinazbabaee]

Dear NiBabies Experts,

I am using NiBabies to preprocess resting state fMRI of infants (2-month-old). I have both T1w and T2w images with the following dimensions, respectively: 3, 140, 256, 256 and 3, 156, 256, 256

I used the following code:

nibabies-wrapper docker /Users/parinaz/Desktop/dat /Users/parinaz/Desktop/dat/nibabies participant --participant-label 23083 --age-months 2 -w /Users/parinaz/Desktop/work/ --fs-license-file /Users/parinaz/Documents/license.txt --skip_bids_validation --use-aroma --md-only-boilerplate --stop-on-first-crash --output-spaces MNIInfant:cohort-1 --fs-no-reconall

The analysis finished successfully. However, the output segmentation and registration in the standard space don't look fine. I attached the results.
segmentation in native space also doesn't look so good, but it is better than the one in standard space.

I want to know if there is any way to improve the result of segmentation and registration in standard space.

Thank you for your consideration,
Parinaz.

[mgxd]

Hi @parinazbabaee - Automated segmentation is one of the trickier steps to get right in nibabies. I have a few suggestions, ranging from easier to harder that you can try:

• It looks like you are not using FreeSurfer (since you included the --fs-no-reconall flag in your command). The fallback is FSL FAST, which from some limited testing is not expected to perform very well in younger subjects due to a lack of tissue contrast. If you remove that and try again, the pipeline will use infant_recon_all for this.
• If you have a few gold-standard labeled participants (preferably around the same age), you could leverage them using the --segmentation-atlases-dir flag to enable running ANTs JointLabelFusion - more detail on this can be found here: https://nibabies.readthedocs.io/en/stable/faqs.html#multi-atlas-segmentation-with-joint-label-fusion
• You could try using a deep-learning tool, such as https://github.com/DCAN-Labs/BIBSnet, to generate these segmentations, and then feed in these precomputed derivatives via https://nibabies.readthedocs.io/en/stable/faqs.html#leveraging-precomputed-results

[parinazbabaee]

Hi Mathias,

Thank you for your kind reply and helpful suggestions. I have already tried using recon-all from the freesurfer; unfortunately, the results didn't improve much. I will try the other two options, which appear to be efficient in fixing my problem.

Best,
Parinaz.