How to apply CanICA on cortical region?

I want to perform group ICA on multiple subjects. I applied multiple masks on each subject such as it only keeps the voxels that are in the cortical area as shown below. When I apply canICA (it returns independent components with few voxels included (which is also shown below). I am suspect that this happens because a mask that the API computes. They didn’t mention anything about what that masks is and what its job is. I read group ICA papers and I didn’t encounter anything related to such masking. Do you have an idea what this mask is?

With which arguments did you call canICA ? Normally, it internally computes a mask that you can look at.