SUIT cerebellar segmentation gives very small lobules

I have been working on cerebellar segmentation of PPMI subjects with SUIT software [SUIT - A spatially unbiased atlas for the cerebellum and brainstem] it is based on SPM12, e.g., DARTEL for registration).
More than 60% of my 117 subjects are giving reasonalble results, but the rest 30% subjcts are giving extremely small segmentation results (lobules in individual space) and the shape looks fine. I have put an issue on the project on GitHub, and more details can be found here [suit_reslice_dartel_inv].
Any one who has experience with SUIT?