Some thoughts on your questions below

1) The "Compute weighted sum BOLD signal within each region" option is recommended when you have a probabilistic ROI, where the ROI file defines a value between zero and one at each voxel indicating the probability of that voxel being part of this ROI. When selecting this option CONN will use those ROI values to compute a weighted average (instead of a simple average) of the BOLD signal within that probabilistic ROI. 

2) The "mask with grey matter" option is recommended when you have relatively large ROIs and you want to exclude, separately for each subject, those portions of the ROI that lie outside of that subject's gray matter mask (in your case with relatively large spherical ROIs this would be a perfectly reasonable option to use)

3) There is no "optimal" radius, it really depends on the level of homogeneity of the area around your MNI coordinates. In general, if the area is relatively homogeneous then a larger radius is preferred as you will get better SNR for the timeseries represented that area, while if the area is relatively heterogeneous then a smaller radius would be preferable in order to avoid spillage from other distinct nearby areas. 

4) In general CONN will extract ROI-level timeseries from the "unsmoothed volumes" (the functional data before smoothing), so the amount of smoothness that you choose during preprocessing does not change the BOLD timeseries extracted within each ROI (it only affects voxel-level connectivity measures). And regarding voxel size, it also generally does not affect your choice of radius, as that choice is often based on the relative anatomy and functional specialization of the area (point 3 above), while the total number of voxels within the ROI only plays a secondary role. 

Hope this helps

Alfonso

Originally posted by Anna Czartoszewska:

Dear Alfonso and CONN users,

I have not had much luck finding detailed information on the ROI-extraction options in CONN. I am working with several spherical ROIs defined from MNI coordinates, including cortical, cingulate, and parahippocampal regions.

In particular, I am curious about how CONN handles potential white-matter inclusion within these seed regions. 

1. What exactly does the option “Compute weighted sum BOLD signal within each region” do? Under what circumstances is it recommended? What are the weights based on - do they somehow relate to segmentation?
   
2. Similarly, should I enable the “mask with grey matter” option, and in what situations is this advisable?
   
3. I am also uncertain about the appropriate radius for spherical ROIs. CONN’s default is 10 mm, but many papers use values between 4-6 mm. Larger spheres include more WM and adjacent structures, which is my concern. 
   
4. Additionally, does the choice of ROI radius depend on the smoothing kernel or voxel size?
   

Thank you in advance for any help.
Best regards,
Anna