We are running longitudinal studies using SCT, and are hoping to contribute to the software.
I’ve noticed there isn’t a dedicated longitudinal pipline for segmentation of CSA and for GM/WM segmentation.
Is there an appetite for this functionality in the community/ is this already being worked on?
Would love some feedback before we dive into this project
First of all, thank you for your enthusiasm in contributing to SCT. You’ll see, the dev team is very friendly
There have been longitudinal studies where SC and GM/WM CSA were measured across time points, for example:
What is typically done, and already possible with SCT, is to compute CSA at each time point. What did you have in mind for the longitudinal analysis?
We have some experience with longitudinal seg. pipelines - essentially creating an intermediate space (template) for the segmentation, so each participant can act as its own control. Then performing the segmentation on the intermediate space and using this as a prior for the individual time-points.
See: https://www.biorxiv.org/content/10.1101/170209v1.full.pdf and https://doi.org/10.1016/j.neuroimage.2020.116798 for examples/details.
We would have to work out the details for the last step, but it should be pretty straightforward. My hope would be reduced within-subject variance across time-points (our cohort move a lot).
Brilliant! Yes, I would be very much interested in seeing such an approach for longitudinal analysis, and helping in any of my capacity. Given that SCT is heavily relying on ANTs, it won’t require additional dependences to the ‘SCT ecosystem’.
Since this looks more like a pipeline (ie ensemble of sequential syntax calls) rather than a single and isolated task, I suggest to develop this pipeline in a separate repository as opposed to creating a dedicated function in SCT, which coverage would be too large. For this purpose, there is a separate organization, called sct-pipeline, which includes pipelines for various analysis scenarios. This could be one of them. My suggestion for the next steps (please feel free to comment):
- Create a repository in the sct-pipeline organization;
- Give your team admin rights to this repos;
- Document in a README.md the overall purpose of the pipeline and its overview (big picture step-by-step);
- Open specific issues for atomic steps to undertake, so we can conveniently assign someone to do it (from your team, my team or someone else willing to help);
- Identify an open-access dataset for testing;