While this paper is cool, I'm not sure it reflects a real scientific advance. O'Craven and Kanwisher showed in 2000 that brain regions that when subjects imagine faces or scenes, face- or scene-selective brain areas are selectively activated (<a href="http://www.pet.au.dk/~andreas/seminars/cog-exp/files/OCraven%20and%20Kanwisher_menta.pdf" rel="nofollow">http://www.pet.au.dk/~andreas/seminars/cog-exp/files/OCraven...</a>). This paper builds on that previous result by showing that this result also holds for hypnagogic imagery. It's interesting but not really unexpected.
Here's the original paper: <a href="http://www.sciencemag.org/content/early/2013/04/03/science.1234330.full" rel="nofollow">http://www.sciencemag.org/content/early/2013/04/03/science.1...</a>
Apart from the populistic BS that the magazines blow all things neuroscience up to, the machine learning technique behind these kinds of studies are actually very clever (and a lot more robust than most "traditional" fMRI findings that produce headlines like "scientists discovered brain area causing teenage angst". If you want a former neuroscientist ranting about bad fMRI research for a page or so, let me know :)