Poor freessurfer mask refinement?

Dear experts,
I am testing different configurations with fmriprep preprocessing pipelines (fmriprep-22.0.2). I first tried one with freesurfer surface processing (–no-submm-recon) and the brain masks looked wrong (too narrow), which then caused inaccurate normalization to MNI. I then switched to --fs-no-reconall and the brain masks and normalization got better. This was observed in more than half of my participants. I don’t know if the brain mask obtained from freesurfer reconstruction made the ANTs worse instead of refining it.

–no-submm-recon

–fs-no-reconall

Thanks,
Oliver