Closed tsalo closed 3 years ago
I'm not sure what my thoughts are on the threshold just yet, I got stuck on the AAL + NS "finger" part. We know AAL is structural, thus not optimal, and we know NS "finger" is good, but also not optimal. I wonder if we could boost signal by also adding + manual finger tapping meta-analysis result to the conjunction. The only one I can think of is Witt et al. from years ago, but I have the coordinates from it, so you could quickly run them through the current version of ALE to get a solid result.
Then perhaps the arbitrary thresholding is less of an issue bc you're using convergence from three different methods: structural + automated (lots of papers) meta-analysis + manual (fewer but more precise paper selection) meta-analysis.
Still doesn't address the hemispheric differences in size tho...
The motor cortex is well constrained to the precentral gyrus, so AAL is at least useful for making sure we don't identify finger-related ROIs not located in M1 (e.g., cerebellum). It definitely isn't good enough on its own, though.
I think the ALE meta-analysis would be great, thank you! I can at least look at the convergence between the two sets of meta-analytic findings and if they don't overlap we can revisit this issue.
I think that varying the thresholds to get equally sized ROIs (as long as the size is chosen a priori) should work. We just need a good ROI size to aim for.
Just sent you the files, let me know how it goes!
Per @handwerkerd's comment on the manuscript, I'm going to look into manually tracing the hand knob.
First draft of process:
Experiment 3 is pretty simple- we apply different denoising strategies to Cohen, Nenka, and Wang's multi-echo finger tapping dataset from OpenNeuro, run a group-level analysis on the despised data, and evaluate the z-statistics from left and right motor cortex ROIs to see how the denoising strategies impact task findings.
The modeling is pretty straightforward, given that filtering is not considered necessary for multi-echo data and that the finger tapping task is a very simple block design. However, I need a principled way to define the left and right motor cortex ROIs. The approach I'm currently considering is to use conjunctions of the AAL left/right precentral gyrus masks and the Neurosynth "finger" meta-analysis (specifically the association/reverse inference/specificity analysis). I can threshold the meta-analysis' z-statistic image using some very high, but arbitrary, threshold (e.g., 10), and then grab the largest cluster from the intersection of that thresholded map and the hemisphere-specific precentral gyrus mask from the AAL atlas.
My concerns are that the threshold is arbitrary and that the resulting ROIs can be differently sized depending on the hemisphere. Is there a more principled approach? If not, I was thinking of using a loop to vary the threshold until the resulting ROIs are the same size, which would have to be decided ahead of time. I'm not sure what ROI size is most appropriate, and I imagine that that would be arbitrary too...