Summary: Why can some memories feel as vivid and detailed as photographs? A new human study has decoded the neural language behind visual imagination. By recording electrical activity from individual neurons in patients with implanted electrodes, researchers found that imagining an object reactivates many of the same cells that fired when the object was originally seen.
This shared biological mechanism explains why mental imagery can seem so realistic and points to new avenues for treating conditions such as post-traumatic stress disorder (PTSD) and obsessive-compulsive disorder (OCD), where intrusive, vivid images cause significant distress.
Key Facts
- Shared neural blueprint: About 40% of neurons that responded while participants viewed an object or face showed the same activation pattern when the participant later imagined that same item.
- Fusiform gyrus and ventral temporal cortex: The study focused on these high-level visual areas. Decoding the neurons’ activity allowed researchers to predict which object a person was imagining from brain signals alone.
- Generative AI validation: The team used generative AI to produce novel images from the discovered neural code and confirmed that these AI-created images elicited the predicted neural responses.
- Clinical potential: Mapping how the brain recreates images suggests ways to reduce the intensity of traumatic flashbacks and intrusive imagery in psychiatric disorders.
- Cross-species confirmation: The human neural code for object representation mirrors findings previously reported in nonhuman primates, establishing a common biological basis for perception and imagination.
Source: Cedars-Sinai
Why do remembered images often feel so real?
A study led by investigators at Cedars-Sinai Health Sciences University provides a clear answer: imagination reuses the very neurons engaged during perception. Published in the journal Science, this research offers the first detailed single-neuron account of how visual perception and mental imagery share a common neural mechanism in humans.
“We form a mental image of a previously seen object by reactivating the same brain cells that responded when we first saw it,” said Ueli Rutishauser, PhD, director of the Center for Neural Science and Medicine and professor of Neurosurgery, Neurology and Biomedical Sciences at Cedars-Sinai, and a senior author on the study. “Our work reveals the neural code the brain uses to recreate images.”
The findings supply a biological foundation for visual imagination, a faculty central to creativity and artistic expression. They also suggest therapeutic targets for conditions where mental imagery becomes uncontrollable or distressing, noted Adam Mamelak, MD, director of the Functional Neurosurgery Program and co-author of the study.
To carry out the experiments, researchers studied 16 adults with epilepsy who had temporary electrodes implanted to localize seizures. Participants viewed a set of images showing faces and objects; a subset later attempted to imagine those same images from memory. During both viewing and imagination tasks, the team recorded electrical activity from hundreds of individual neurons in each patient.
Neurons in the fusiform gyrus—an area critical for high-level visual processing, especially face perception—showed selective responses to specific image features. For roughly 80% of visually responsive neurons, researchers could identify the image aspects that drove each neuron’s activity, revealing a clear neuronal “axis code.” When participants imagined the objects, about 40% of those axis-tuned neurons reactivated using the same code, reproducing the activity pattern seen during perception.
“Advanced artificial intelligence tools were essential throughout,” said Varun Wadia, PhD, a postdoctoral scientist and first author. “We used deep visual neural networks to convert objects into numerical descriptions, which let us decode neurons’ tuning. Then we validated the code by generating new images with AI and predicting the brain’s responses to those novel stimuli.”
The study builds on prior work by Doris Y. Tsao, PhD, of UC Berkeley, who identified a similar neural code for object recognition in macaques; Tsao is a co-senior author on this human study. Together, these results indicate the same axis-based representation of visual objects exists across species and underlies both seeing and imagining.
Funding and commentary highlighted the study’s broader implications. “These findings support the idea that imagining and seeing share a common neural code and may help us understand psychiatric disorders involving disrupted imagery and reality monitoring,” said Hermon Gebrehiwet, DrPH, program officer at the National Institutes of Health.
Open questions remain, including what internal signals initiate the reactivation and how memories selectively recruit the precise subset of neurons needed to recreate a target image.
Other Cedars-Sinai authors: C. M. Reed, J. M. Chung, and L. M. Bateman.
Funding: Supported by the NIH BRAIN Initiative (U01NS117839 to UR), the Howard Hughes Medical Institute (DYT), the Simons Foundation Collaboration on the Global Brain (UR and DYT), and the Chen Center for Systems Neuroscience at Caltech (DYT).
Key Questions Answered:
A: The underlying code is the same, but imagination reactivates fewer neurons and with weaker overall responses than actual perception—about 40% of the neurons that fired during sight. This reduced activation acts like a volume control, helping the brain distinguish internal images from external reality. Conditions such as schizophrenia may involve failure of that control.
A: Progress is being made. By combining deep visual networks with neural recordings, researchers can now infer object categories people are visualizing with increasing accuracy. However, we are not yet able to reconstruct continuous movies of thoughts or dreams.
A: The results suggest a biological basis: people with exceptionally vivid imagery (sometimes called hyperphantasia) may reactivate a larger proportion of visual neurons during imagination, making internal images feel more like perception.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional context was added by the editorial staff.
About this research on hyperphantasia and visual neuroscience
Author: Cara Martinez
Source: Cedars-Sinai Medical Center
Contact: Cara Martinez – Cedars-Sinai Medical Center
Image: Image credited to Neuroscience News
Original Research: Closed access. “A shared code for perceiving and imagining objects in human ventral temporal cortex” by V. S. Wadia, C. M. Reed, J. M. Chung, L. M. Bateman, A. N. Mamelak, U. Rutishauser, and D. Y. Tsao. DOI: 10.1126/science.adt8343
Abstract
A shared code for perceiving and imagining objects in human ventral temporal cortex
INTRODUCTION
Mental imagery is the brain’s ability to produce percepts, emotions, and thoughts without external input. It supports creativity, memory, planning, and simulation of actions or outcomes. When imagery becomes uncontrollable, it contributes to psychiatric conditions such as anxiety, schizophrenia, and post-traumatic stress disorder. Despite its significance, the single-neuron basis of visual imagery in humans has been unclear.
RATIONALE
To investigate single-neuron mechanisms of imagery, the team recorded individual neurons in human patients implanted with diagnostic electrodes. They focused on the ventral temporal cortex (VTC), a region specialized for representing visual objects. First, they determined how individual neurons encode visual objects; then they tested whether that same code reappears during imagination.
RESULTS
Across 16 patients, researchers recorded 714 neurons in human VTC while participants viewed objects. A majority (456 of 714) were visually selective across five categories: faces, plants, text, animals, and objects. Using activations from deep networks to build a low-dimensional object space, the team found that nearly 80% of visually responsive neurons were tuned to specific axes in that space. In a subset of patients who also imagined the objects, about 40% of those axis-tuned neurons reactivated during imagery, and their responses during imagination were proportional to how strongly an object projected onto each neuron’s preferred axis. This allowed reconstruction of imagined objects while still distinguishing imagined from perceived stimuli.
CONCLUSION
Recording from the same human VTC neurons during perception and imagination shows that visual objects are represented with an axis code and that imagination reactivates this code. These single-neuron results support a generative model of mental imagery and offer new directions for understanding and treating disorders tied to intrusive visual imagery.