general-discussion > RestvsTask: runs of unequal length
Showing 1-2 of 2 posts
Results per page:
Jul 17, 2013  08:07 AM | Maxime Pelland
RestvsTask: runs of unequal length

I have acquired BOLD time series while participants were resting and while they were participating in a task. The runs for the rest are much shorter (less volumes) than the ones of the task. I was wondering whether this would create an important bias if I launch niak_region_growing in the data set. That is, will the segmentations more representative of the task than the rest. 
Also, I will be using niak_pipeline_stability_rest to obtain partitions of the brain at different scale. Will there be a bias here due to the unequal length of the task and rest time series?

Jul 17, 2013  09:07 AM | Pierre Bellec
RE: RestvsTask: runs of unequal length
Dear Maxime,

You are right about the region growing. It is applied on a concatenation of all functional data across both runs and subjects. So if one type of run is longer than another, you will have a bias at this stage. However the region growing is used to generate small regions, which have little impact on the end result of the clustering. So I don't think this is much of a concern. 

In BASC, you have the possibility to specify "strata" of subjects. That's done by feeding a csv file in (see help niak_pipeline_stability_rest). If you code each run as a different subject, you can define one stratum "rest" and one stratum "task". Both strata will be given equal weights in the group analysis. But this is a tweak of the original design of the pipeline, that will require to code manually the structure files_in instead of using the grabber. I wouldn't go through the trouble of implementing that unless you have a very solid reason for wanting an unbiased functional network decomposition. Alternatively you can try to run BASC on task and rest only, as well as a consensus of the two, and check whether this matters for the final statistical analysis you are running on your data. 

I hope this helps. Best,