Dear experts,
first of all, thank you very much for developing this valuable toolbox and sharing it with the community — I can imagine that it involves a lot of work and effort.
Now to my questions: I am planning to perform a data-driven
parcellation of the striatum based on a resting-state fMRI dataset,
as described in Hinkley et al., 2023 (https://doi.org/10.1016/j.schres. 2023.08.030), which is
based on Jung et al., 2014 (https://doi.org/10.1371/journal.pone.010...) and Kahnt
et al., 2012 (https://doi.org/10.1523/JNEUROSCI. 0257-12.2012) to test
the intrinsic activity of the subregions of the striatum identified
in this way for correlations with several questionnaire measures.
After intensive research in this forum and taking into account the
functions added to CONN over time, I came up with the following
pipeline to perform the parcellation:
1. Preprocessing and denoising as described at https://web.conn-toolbox.org/tutorials
2. At the 1st-level perform similarity (RSIM) (is this the 'RSC' measure suggested by Alfonso in his post https://www.nitrc.org/forum/message.php?... to focus on?
3. At the 2nd-level, define the contrast [1 0 0;0 1 0;0 0 1] as suggested by Alfonso at https://www.nitrc.org/forum/message.php?... click on the 'display results' button and select 'SPM display.'
4. [From this point on, I am increasingly unsure whether the
procedure is correct] In SPM (I'm using SPM12), select the contrast
'connectivity result', apply masking 'image,' select a mask (I was
thinking of combining left and right caudate and combining left and
right putamen from the Neuromorphometrics atlas, implemented in SPM
into two separate masks; is this atlas fine, or can you recommend
any other atlas for these specific regions?), nature of mask
'inclusive', right-click in the SPM glass-brain 'go to global
maximum', right-click in the SPM glass-brain 'Extract data' >
'whitened and filtered y' > 'This cluster', save Variable y in
MATLAB Workspace for use in step 6.
Note: In the posts https://www.nitrc.org/forum/message.php?... and https://www.nitrc.org/forum/message.php? msg_id=9874, it
is suggested to use the SPM function 'small volume,' but it seems
to me that this only adjusts the table; the voxel-values still
cover the whole brain. Therefore, I use implicit masking here to
obtain the values I need for k-means clustering to parcel the
striatum. Or am I getting something wrong?
5. Repeat 4. with the other mask
6. Perform K-means clustering on each saved y from step 4 separately for several K's (2 to 10)
7. Determine the optimal number of clusters
8. Extract mean intrinsic activity from the identified optimal clusters/sub-regions (or better small spheres at maximum coordinates of the clusters instead, as in Hinkley et al., 2023?) for testing correlations with questionnaire measures
Is my planned approach correct or have I misunderstood something, failed to take something into account, or are there more elegant solutions for data-driven parcellation of the striatum?
Many thanks in advance,
Mike