Your brain is now their open book [UPDATED]

Oops.  [Much obliged, Pete.]

Scientists claim to have extracted images directly from the brain.  They trained people to read the word "neuron" with their eyes really really hard until the scientists' local decoders decoded a visual image reconstruction of their subjects' retinotopic map that displayed the following multivoxel pattern of fMRI and multiscale visual representations:

The mind is now an open book!  These people not only have neurons in their brains—they have electrical images spelling out the word "neuron" on their brains.  The signifier is become the the signified!  All praise whichever one you think is more important!

But please ignore the fact that the "n" at the beginning of that first "neuron" looks like this:

Whereas the "n" at the end of the word "neuron" looks like this:

Or that the "n" at the beginning of that second "neuron" looks like this:

Whereas teh "n" at the end of the second "neuron" looks like this:

Seems the human brain represents letters on itself so precisely that the algorithms these scientists used to train local decoders to select relevant voxels and assign weight matched signifier to signified down to the last pixel. 

Or some scientists just discovered Photoshop and thought they'd frighten us with visions of regimes whose fascism was founded on dream-reading devices.  I'm banking on the former.  (Wim Wenders?  He's already banked on the latter.)   

Comments

[Pete Skomoroch]

These were not photoshopped. The legend for the figure in the paper explains that the last column is from the same data as the first column (five unique letters were presented randomly multiple times in the "figure image sessions", the letters 'e','u','o','n' and 'r').

From the legend:

"...Each reconstructed image was produced from the data of a single trial, and no postpro-cessing was applied. The mean images of the reconstructed images are presented at the bottom row. The same images of the alphabet letter ‘‘n’’ are displayed in
the rightmost and leftmost columns for each subject."

They didn't train on the word neuron, or the individual letters. The decoding algorithm was trained on fmri data from images of random shapes, which then predicted the letters given fmri corresponding responses.