McNemar works only on the subjects that have different classifications so the sample size of 2972 is really irrelevant, as is the 63 for the skull data.
Can you provide the data in this form for both sets. Only the yes/no and no/yes cells are actually important.
I don't really understand the setup. However, to use McNemar you need two yes/no type responses for each subject from two claimed equivalent treatments. It doesn't look like that for the MRI.
The mention of MRI and other medical stuff indicates that you may be dealing with serious clinical situations. If that is the case, my strong advice is that your department buys an hour of a statisticians time who is prepared to sign off on any conclusions you make. You can hardly quote advice from some random guy on the internet.
Good luck with that. kat