AI can recreate images from brain activity with up to 75% accuracy

Previously, recreating images from brain activity was only possible when the subject was looking at the images, or the images—like faces, letters, or simple drawings—were very clear. A team of researchers from the National Institute of Quantum Science and Technology (QST) and other institutions has demonstrated that it is possible to recreate all kinds of images, like landscapes or complex drawings, to some extent based on thought alone.

The team first used MRI scans to record the brain activity of subjects who viewed 1,200 different images. They then used AI to create a “dot map” with about 6.13 million elements, such as colors, shapes, and textures. A neural signal decoding program was created to match brain activity to the dot map, generating new dot maps as new brain activity was input.

Next, the subjects were shown another image of the 1,200 they had seen before, and their brain activity was measured by MRI 30 minutes to an hour later, while they were asked to visualize the type of image they had seen. As the recordings were imported, the neurosynthetic decoder generated a score plot. The plots were fed into another synthetic AI program to reconstruct the image, undergoing a 500-step revision process.

This process helps identify the original image from the reconstructed image with an accuracy rate of up to 75.6%, a significant improvement over previous methods that achieved only 50.4% accuracy. The results are published in the online edition of the international scientific journal Neural Networks.

The research could lead to new forms of communication without using words. "It's a huge achievement to be able to see into the minds of others for the first time," said QST researcher Kei Majima. "I hope this research will lead to a deeper understanding of the human mind."