Hi,

If you want to calculate the sample size for future data analyses you could use "fmripower" (http://fmripower.org/; http://fmripower.org/instructions.pdf). You may need some pilot data. Posthoc power analysis are nor recommended (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3427872/).

Regards

I looked at the Neuroimage paper, though it seems to be geared towards powering task based fMRI. As the methods for obtaining resting state results are fundamentally different, I don't think you will be able to use fmripower for that. This is a good question, as I am not sure that a clear answer exists. Maybe a manuscript on the subject is warranted. Feel free to contact me if you'd like to discuss further. 

@ Roberta & Matthew: I am not really familiar with fMRI analysis, but as the recommended paper calculates Cohen's d, what is the difference between fMRI power calculation and any other power calculation?

This is an excellent question which has not been addressed in the literature yet.

In the fMRI with task, this question was addressed by Thirion et al (10.1016/j.neuroimage.2006.11.054) and found that 20 to 25 subjects were necessary to achieve reliable results. In the resting-state literature, as there is no task, I personally think we need at least to double the number of subjects in each group, that is 40 to 50 subjects per group if possible. This explains the huge effort made by the neuroimaging community to share such data ! 

You additionally need to acquire scans with long acquisition duration (at least 12 min) as reliability increases with duration as shown by R. Birn and co-authors (http://dx.doi.org/10.1016/j.neuroimage.2013.05.099).

The preprocessing and analysis pipeline you'll use will be of importance as demonstrated by Shirer et al. (http://dx.doi.org/10.1016/j.neuroimage.2015.05.015).

Hope it will help !