Dear Agurne,

Both are good and tricky questions.

Regarding (1) there is no agreed-upon threshold of when TR is "sufficiently" low to warrant skipping slice-timing correction, I would say that if you are focusing on low-frequency fluctuations (e.g. below 0.10 Hz) then 1.121s is likely already low enough (since most of the periods of the BOLD signal you will be focusing are considerably larger than 1s), but if you are interested in analyzing high-frequency fluctuations, and/or if you are focusing on task-related effects as opposed to pure resting-state scans, then you probably would still benefit from including slice-timing correction. 

Regarding (2), I typically recommend using the most conservative threshold that your data can afford, so basically try the 95th percentile first (which is a bit more conservative), and see if that works fine (i.e. if that does not remove too many scans, which you can determine during the Denoising step by making sure that the residual degrees of freedom after denoising are "sufficiently" large, e.g. above 50). There are many cases where this more-conservative option will likely not work, like in children or people with larger-than-average motion within the scanner or in tasks that involve moving or speaking within the scanner, and in those cases you may need to revert to the more liberal 97th percentile option, or even the more liberal 99th percentile option and attempt to further control for any potential residual motion in posterior steps (e.g. in second-level analyses)

Hope this helps
Alfonso
Originally posted by Agurne Sampedro: