Hi there. I’m a research assistant at the VCU Wellbeing Lab. We’ve recently been interested in adapting your ISC and ISRSA tutorials for use in fNIRS research. I have managed to some feed data through the pipeline, calculating ISC stats, p-values, and null distributions for every fNIRS channel (24 channels) instead of the tutorial’s fMRI analysis for every ROI. Now that I have done so I’m not entirely sure if it is valid because the underlying nltools functions are perhaps only intended for fMRI and the math breaks? Or maybe it doesn’t? The nltools documentation for the nltools.stats.isc function looks ambiguous. Double thanks in advance, I really like the book and the math!
Hi @orionpearce1984, thanks for your questions. The ISC functions in nltools should generalize to most other types of data. We use it with fMRI, intracranial, psychophys, face expressions, etc. I can’t think of what else you might want to consider that might be specific to fNIRS. One thing to note is that I don’t recommend focusing too much on the dynamic ISC function yet as there are several other ways to compute that metric that give different responses and we haven’t implemented them all yet.
I would say the most important things to consider are:
-
What distance/similarity metric are you interested in ? (pearson, spearman, cosine, euclidean etc) they each have different implications
-
Do you want to compute the mean or median of the similarity/distance matrix. If using the mean, make sure you do the fisher r to z transformation before computing the mean. (this is done by default in nltools and brainiak).
-
How do you want to perform a hypothesis test? Subject-wise bootstrapping? Circle shifting or phase scrambling?
If you haven’t read this yet, I highly recommend starting with the practical tutorial on the naturalistic-data.org jupyter-book. <edited: sorry just saw you posted this in the naturalistic-data.org channel - my bad!>
Good luck!