GoKawiilWeizmann Institute's Brain-IT AI reconstructs images from fMRI brain scans
Researchers led by professor Michal Irani at the Weizmann Institute of Science built an AI model called Brain-IT that reconstructs images a person was viewing based on functional MRI data, and can also predict what a brain scan would look like for a given image. The team reports it needs only about one hour of fMRI data from a new subject to match results that other methods achieve using 40 hours of recordings, and claims improved accuracy on composition, color and detail compared to prior models. The work has been shared on GitHub and presented at a scientific conference.
GoKawiil's interpretation of the reporting above, not reported fact.
The dramatic reduction in required scan time could make brain-decoding research more practical to scale beyond small lab studies, since fMRI time is expensive and subjects can struggle to sit still for long sessions. Irani and colleagues suggest potential medical applications, such as helping people with injuries or disabilities communicate, though such uses remain speculative and not yet demonstrated clinically. The researchers also raise the possibility of eventually decoding dreams, an ambition that goes well beyond what the current results show.
- Brain-IT reconstructs viewed images from fMRI brain scans with notably higher fidelity than prior AI approaches, according to its developers.
- The model reportedly needs only 1 hour of a new subject's fMRI data to match results that competing methods achieve with 40 hours.
- Researchers point to possible future uses in medical communication aids and, more speculatively, dream decoding.
Source: cnet.com — Joe Hindy, 2026-10-08
Published there as: “This ‘Mind-Reading’ AI Is a Wiz at Figuring Out What You See”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.