Scientists from the Kamitani Lab of the Japanese University of Kyoto have developed a neural network that has been able to read what a person has in their mind and, in addition, to reflect it on a screen.
They have achieved this through an algorithm that interprets and accurately reproduces the images recorded by the brain after viewing them on a screen. He is also able to interpret and reproduce what a person remembers from the images he has seen.
This is a technological feat, since until now the results obtained by using algorithms to decode mental images have been very limited. The new work has not only increased the accuracy in the perception and computer reproduction of mental images, but has also been able to interpret and replicate forms that are only in the imagination of a person.
Although the possible applications of this new technology are not immediate, this research represents an important step forward to build systems that can help us project the interior of our mind to the outside of the brain, says the journal Science.
The research involved three volunteers with normal vision, who were presented with three different categories of images: nature, letters and geometric shapes.
When contemplating these images, an activity of the cerebral cortex was generated in the brains of the volunteers, which analyzed the neuronal network. The volunteers observed more than 1,000 images several times, including a fish, simple color shapes and an airplane, stimuli that the neural network managed to interpret correctly.
To discover what a person is seeing, researchers first turned to functional magnetic resonance imaging (fMRI), which measures the blood flow in certain regions of the brain and thus reveals neuronal activity.
Through the fMRI they observed the cerebral areas of visual processing corresponding to each image. They analyzed the brain activity before the response to each one of the images separately, to get the computer to paint a reconstructed image from interpreting the brain activity. The reconstruction of the image was achieved through a deep neural network created especially for this research.