When i manually have corrected some areas of subcortical structures (ex Caudate) i still get some pixels left of other regions (NAcc for ex)  in the corrected area after I again have run recon-all -autorecon2 and -autorecon3. 

It seems like the program does not read my alterations correct.

Why is this? Is there anything i can do to prevent this? 

I have off course saved segmentation after my corrections before i re-run them.

Kind regards, 

Elin Allzén