Weizmann’s Brain-IT rebuilds viewed images from fMRI scans with one hour of calibration
Researchers at Israel’s Weizmann Institute of Science developed Brain-IT, an AI system that reconstructs images people view using activity recorded by fMRI scans. According to MIT Technology Review, the system separates predictions of an image’s structure from its content, and needs about one hour of data from a new person, compared with roughly 40 hours for earlier approaches.
Bottom line — Brain-IT cuts calibration to one hour, but its researchers have not yet shown it can reconstruct imagined images.
Go deeper 7
-
MIT Technology Review says the team trained its system using data from eight people, each shown around 9,000 images in a high-resolution scanner.
-
The same report says about 70% of the training images were not originally paired with fMRI scans.
-
The encoder predicts brain activity from an image; the decoder uses brain activity to reconstruct one.
-
The system can still misread scenes: MIT Technology Review reports a cake rendered as three sandwiches and a dog in a bathtub as a goat.
-
Researchers hope the work could eventually support communication for people who are locked in, but reconstructing imagined images remains a goal, not a demonstrated capability.
-
Tommy Sprague, a neuroscientist at the University of California, Santa Barbara, told MIT Technology Review that fMRI imaging costs around €550–€920 an hour ($600–$1,000).
-
Sprague and neuroscientist Marcello Ienca, of the Technical University of Munich, warned that future brain-decoding tools could raise mental-privacy concerns, particularly if adapted to EEG.