The brain activity was then put into a computer program. During a second round, Kay and Naselaris were then shown 120 different photos. The computer model sifted through its previous store of brain activity-to-image patterns to "decode" the second round of fMRI data and find a correct match.
The authors stressed that their decoder program was not attempting image reconstruction, but rather image identification.
The results: When given a set of 120 photo options, the computer successfully identified the viewed images between 72 percent and 92 percent of the time. Broadened to 1,000 images, the success rate was 80 percent. With a pool of 1 billion images (as many, they noted, as are cataloged online by Google), the authors estimated that the decoding model would work about 20 percent of the time.
Gallant discussed a number of ways in which a fully developed method for decoding brain imagery might ultimately be applied as a practical medical tool.
"In theory, this could be used to help doctors evaluate the effectiveness of drugs designed to improve brain function," he noted. "Or it could, perhaps, be used to help fit neuro-prostheses for the blind, or to assist with psychotherapy, the interpretation of dreams or biofeedback. But all this is a long way down the road."
Dr. Joe Verghese, an associate professor of neurology at the Albert Einstein College of Medicine in New York City, agreed that the study has interesting implications but described it as "just a first step" in a complex effort to decode the brain.
"This kind of pattern recognition -- while consistent with previous work -- is still very targeted, very defined and very crude," he said. "They're using the noninvasive beauty of the MRI to look at the part of the brain that deals with vision, rather than trying to read thoughts. But I can see how, if they can carry it forward, there could eventually be medical applica
All rights reserved