The images look like Stable Diffusion riffs based on an example image.
Might it be that they are doing something like this:
* presenting some images to a subject, converting them to some captured MRI signals.
* training an A.I. image generation system on those same images, using the MRI signal as the associated text, instead of text.
* then generating images using MRI patterns as prompts
Still, scary what the method is able to infer.