Sounds like something straight out of a futuristic science fiction movie, but the technology is being used now-researchers can decode the brain to see what a person sees.

Researchers at the UC Berkeley Helen Willis Neuroscience Institute have developed a technique that can predict what images an individual is viewing by using brain-scanning equipment and computer models. The study was published in the March 5 issue of Nature science journal.

Potential advancements derived from this method could be applied to various fields. In psychotherapy, it can be used to interpret dreams, and in medicine, it can help the blind see again or aid physicians to

better diagnose brain damage caused by stroke.

"The goal of this research is to obtain a quantitative computational model for the human visual system," investigator Jack Gallant, an associate psychology professor at UC Berkeley, wrote in an e-mail. "This is one of the critical steps necessary to understand the visual system (and other areas of the human brain), and is a prerequisite for designing any sort of treatment."

The model relies on functional magnetic resonance imaging, which traces minute blood flow in the brain. By analyzing primarily the blood flow patterns in the visual cortex, the program can tell researchers which pictures were shown.

Previous brain-decoding methods were limited in that they could only pick up on images that the individual had seen before-Gallant's model can describe novel images never previously viewed by the subject.

To discover this technique, two of the study's co-authors looked at 1,750 different images depicting black and white photographs of natural scenes, "just like those that you would take on vacation," Gallant said.

Researchers then displayed 120 new images never seen before by the subjects. The team discovered that they could accurately predict which picture was shown to one subject 92 percent of the time.

However, the more novel images in the visual repertoire, the less certain the computer model becomes. When the image stock pile was increased to 1,000, the accuracy declined to 82 percent. The campus researchers estimate that if a billion pictures are used, around the number of images catalogued online by Google, accuracy drops to 20 percent.

The scientists also note that the model allows for image identification, not image reconstruction. That is, the program can only determine what a person is looking at from a known set of pictures.

"Our experiment solved the identification problem, no one can do reconstruction yet," Gallant said.

The Gallant lab on campus primarily investigates vision and the neural mechanisms that allow for visual perception.

"Vision is a good subsystem of the brain to work on because the input (of the visual stimulus) can be easily controlled and measured," Gallant said. "Because the brain is built on modular principles anything we learn about one brain subsystem generally applies to other systems as well."

Further plans for research include creating more models for other areas of the brain involved in vision to possibly improve accuracy of identification of images.

"There are probably several dozen distinct areas in the human visual system, and we currently only have good models for two of them," Gallant said. "So there is lots of work to be done."

Tags: VISION, PSYCHOLOGY