Block Averaging/Repeated Measures vs. GLM
Hi David and Meryem,
We briefly discussed our dataset last week during the live session of the fNIRS training course, but I have a follow up question that I'd like to ask here.
We have collected fNIRS data from 60 infants (10-months) whilst they completed a touchscreen task that required them to manually respond to the screen. The task is divided into blocks (control and experimental) separated by a baseline video. The baseline duration is jittered (10-17s) and the duration of the blocks varies each time, and for each participant. This is because there are 6 trials in each block, and the block only ends once the participant has completed the 6 trials. Block durations vary from around 25s to over 90s. We have used Homer to pre-process our data.
My question relates to how to statistically analyse this data. Would we be able to run the GLM in Homer3 (as you've shown us last week) if we have varied block durations, no SS channels and no auxiliary measurements. If we are able to, what would be the advantage of running the GLM? For example, would their be advantages of using the GLM over the block averaged approach and running repeated measures analyses to identify channels showing sig changes (from baseline) in oxy and deoxy Hb in experimental vs. control conditions?
There are some advantages to using GLM instead of block averaging even when you don't have short separation measurements or other auxiliary measurements. These include:
- you can use temporal basis functions to model your HRFs
- you can better model variable trial lengths. This is not yet implemented in Homer3 but it will be very soon
- collectively, 1 and 2 should provide you better statistics compared to block averaging
- you can incorporate drift correction into the GLM instead of having to do a preprocessing band pass filter as is required with block averaging. The benefit here is that it reduces the changes of the band pass filtering removing part of your brain activation if it is incorporated into the GLM to simultaneously estimate the HRF
There are no doubt other advantages as well which I hope others will bring up in this thread
Adding in variable trial lengths in the GLM is a high priority item for us that Stephen and Jay want to finish ASAP. Please feel free to check back here and ask us a week from Friday to see if it is done yet.
@afiske, it is not ready yet. @stucker is actively working on it. Should be done soon.
@dboas Hi David,
I am a new user of Homer3. I have three questions and I really appreciate it if you can answer them.
1- for the purpose of pre-processing fNIRS data, we should use just one motion correction algorithm for example PCA? or
is it possible to use several motion correction algorithms?
2- after pre-processing, we have to choose between block-averaging analyzing and GLM? or we can use both of them to get the HRF data?
3- for statistical analysis, we have to export the HRF data into a statistical analyzing system?
I really appreciate your help and response.