Attention Freesurfer Users,
If you are using v5.1, you should be aware of a new function in this version that wasn't fully documented in the Release Notes until recently.
Beginning with v5.1, control points (added to fix white matter intensity issues, per normal procedures) are now used in the mri_ca_normalize stage which precedes the creation of the aseg. This functionality was added to v5.1 to help the aseg creation. However, we have discovered that it is possible for this normalization, making use of the control points, to adversely affect the subcortical segmentation, such that some structures differ in size when compared to processing without control points being used by mri_ca_normalize. This effect seems to occur only in instances where adding control points to regions where signal drop-off within white matter is dramatic (even while correcting the white matter intensity issue as intended). We did not observe this affect within our own test sets.
The advice we give is to be sure that control points are not placed into regions which share or near voxel intensity values close to that of grey matter. If a control point is placed deep within a white matter strand (generally not recommended), an additional control point should be placed nearby outside of the strand (within white matter) to limit the spatial extent of the influence of those control points.
An option is to disable the running of mri_ca_normalize when re-running the -autorecon2 or -autorecon2-cp stage after adding control points by adding the flag -nocanorm to the end of recon-all.
The Freesurfer Team