NiftyNet should be able to connect to an XNAT instance to fetch data

[tvercaut]

As discussed with @atbenmurray, rather than alway working from images "dumped" in a folder, it would be great for NiftyNet to support training from data stored on an XNAT server. This will pose a number of questions related to authentication and IO bottlenecks which in the long run would need to be solved.

As a side note, I am referring to an old issue on cmiclab, #205 which had some interesting comments on IO efficiency when working with 2D images.

[ericspod]

This is essentially a duplicate of #183.

[tvercaut]

This is actually not really a duplicate. #183 is about one-off downloading through a command-line whereas #184 is a tighter integration at the API level.