Hi Abir,
Given that you are reporting that you are getting the same value in mrstats's “count” output across multiple subjects, then my best guess is that you are merely providing the mask images to mrstats without using the -mask option. In that instance, what mrstats is telling you is the total number of voxels in the image FoV; not the number of voxels that contain a non-zero value. Passing the same image to mrstats via the -mask option will compute summary statistics within only those voxels present in the mask. Alternatively you could use the -ignorezero option, which will have a similar effect of excluding voxels absent from the binary mask image from the statistics calculation.
Rob