I have some unexpected results from the mne_watershed_bem script that uses Freesurfer mri_watershed program.

For one of my subject's the inner scull surface ( xxxx_inner_scull_surface in bem/watershed/) partially excludes cerebellum (the left screen shot). All my other subjects have cerebellum included into the inner scull (right screen shot). As a consequence later stage scripts (mne_setup_forward_model and mne_do_forward_solution) produce similarly erroneous inner scull surface delineation. 



Did anyone experience this problem before, and if so, how would you advise to fix it?
 
For this subject MRI MP RAGE sequence was acquired using the 12 channel coil, which, compared to the 32 channel one used for other subjects, produced images of worse quality. However, I have other subjects with 12 channel coil scans and their inner scull surfaces were separated normally (including the cerebellum).

Thanks,
Ana