Hi Alfonso and CONN users,
I'm analyzing a pre-post tDCS treatment cohort (n=33) using MVPA to identify voxels with significant FC changes compared to baseline. Each subject undergoes 3 sessions with anodal, cathodal, or sham stimulation applied to either the midline or right lateral cerebellum for all 3 sessions. Post-hoc SBC and RRC analyses are used to further examine MVPA seeds, focusing on directionality and RSNs based on treatment type.
In one contrast (midline Anodal vs. Sham), we found a robust MVPA cluster in the WM area near the thalamus and caudate tail. It's also close to the lateral ventricle which is generally an area susceptible to motion, which got me a little concerned that something failed in preprocessing. I checked single slice montages for structural and functional artifacts and found no abnormalities. QA reports show no significant outliers for PVS, invalid scans, or motion, with metrics above the 95% NH threshold. I also analyzed raw scrubbing values by condition with the idea that there might be more motion during active tDCS compared to sham, and found no significant motion differences between the conditions. Standard pipeline settings were used (8mm smoothing kernel, intermediate outlier detection, etc.).
Post-hoc SBC analyses also shows it is associated with the cortex (visual, IFG, even caudate which is interesting).
With all this said, it seems that everything was preprocessed just fine. I have tried MVPA with 3 and 4 components, and the cluster persists. I'm not sure how often WM activation is reported in MVPA, but I'm wondering if there's something I've overlooked or should try out (perhaps changing some parameters at first level)? I’d hate to change the standard settings just to try to remove the WM finding when there are no preprocessing issues. Given the cerebellum's strong connections to the basal ganglia, there might be some theoretical support for this result. Any insights would be appreciated!
MVPA WM cluster
Post hoc SBC results using WM cluster as a seed
Hi Joseph,
That's an interesting question. I agree with you that the results seem more likely to reflect nearby caudate or PO activation rather than noise effects, particularly given the repeated-measures nature of your analysis (comparing the two conditions) which makes it much less likely to see effects caused by motion differences across subjects or other similar artifactual effects. Also perfectly reasonable are all of the supplementary analyses that you performed, ruling out potential task-by-motion interactions and showing reasonable connectivity difference maps with this cluster, which again point to this being likely a true effect. I would go ahead and report these results as they are. In general the anatomical precision afforded by MNI normalization is not sufficient to resolve this sort of differences without using more sophisticated methods so I think it is perfectly reasonable to report these results as they are and leave to future studies/analyses the question of what is the exact anatomical source of these effects. That said, if you are looking for some additional ideas of potential ways to further explore that question, I imagine that using subject-specific ROIs (e.g. created using freesurfer subcortical parcellation) would be a reasonable way to achieve higher spatial specificity. Also performing post hoc analyses that look at the connectivity with that cluster but restricted to subsets of subjects based on anatomical markers (e.g. subjects where that location appears to more clearly coincide with the tail of the caudate) may also be helpful to characterize functional-anatomical variability associated with these results.
Hope this helps
Alfonso
Originally posted by Joseph Dust:
Hi Alfonso and CONN users,
I'm analyzing a pre-post tDCS treatment cohort (n=33) using MVPA to identify voxels with significant FC changes compared to baseline. Each subject undergoes 3 sessions with anodal, cathodal, or sham stimulation applied to either the midline or right lateral cerebellum for all 3 sessions. Post-hoc SBC and RRC analyses are used to further examine MVPA seeds, focusing on directionality and RSNs based on treatment type.
In one contrast (midline Anodal vs. Sham), we found a robust MVPA cluster in the WM area near the thalamus and caudate tail. It's also close to the lateral ventricle which is generally an area susceptible to motion, which got me a little concerned that something failed in preprocessing. I checked single slice montages for structural and functional artifacts and found no abnormalities. QA reports show no significant outliers for PVS, invalid scans, or motion, with metrics above the 95% NH threshold. I also analyzed raw scrubbing values by condition with the idea that there might be more motion during active tDCS compared to sham, and found no significant motion differences between the conditions. Standard pipeline settings were used (8mm smoothing kernel, intermediate outlier detection, etc.).
Post-hoc SBC analyses also shows it is associated with the cortex (visual, IFG, even caudate which is interesting).
With all this said, it seems that everything was preprocessed just fine. I have tried MVPA with 3 and 4 components, and the cluster persists. I'm not sure how often WM activation is reported in MVPA, but I'm wondering if there's something I've overlooked or should try out (perhaps changing some parameters at first level)? I’d hate to change the standard settings just to try to remove the WM finding when there are no preprocessing issues. Given the cerebellum's strong connections to the basal ganglia, there might be some theoretical support for this result. Any insights would be appreciated!
MVPA WM cluster
Post hoc SBC results using WM cluster as a seed