nailing very atrophied brains to MNI can also be a bit dangerous – it’s a typical bias-variance problem (using a normalization approach that has high variance can lead to getting ventricles to fit but curly/smeary overfitting of smaller regions).
So in any case you should definitely check your data (e.g. take a look at the STN specifically in case of STN DBS) if you use a high variance normalization scheme.
I can recommend the ANTs preset “Effective (high variance)” as a first try.
Prelim data suggests that lower variance schemes do a better job at normalizing the STN though. So be careful with that.
And yes you need to upgrade for that normalization scheme.