[Freesurfer] bad aseg but WM and pial surfaces look good

Hi Freesurfer developers,

I've processed my sample of T1 scans through recon-all and have corrected the pial and WM surfaces. The surfaces look mostly good, though when I checked the aseg segmentations they are pretty poor and there is huge variability in the volumes I get from asegstats2table (for instance, Left-Lateral-Ventricle is anywhere from 3773.1 to 22575.9). I was wondering if there's any way to improve the subcortical segmentations? Moreover, if the subcortical segmentations are poor, should I worry that there might also be errors in my cortical thickness or local gyrification index calculations (though the values for these show much less variability).

Many thanks, Colleen