A Landmark Study from Cedars-Sinai Reveals the Brain's Unified Code for Seeing and Imagining
In a breakthrough published in the prestigious journal Science on April 9, 2026, researchers at Cedars-Sinai Medical Center have provided the first single-neuron evidence that the human brain employs the same neurons to both perceive visual objects and conjure mental images of them. This discovery bridges a long-standing gap in neuroscience, demonstrating how visual perception and imagination share a common neural mechanism in the ventral temporal cortex (VTC), a key region for object recognition. The study, led by postdoctoral scientist V. S. Wadia and senior authors Ueli Rutishauser, PhD, and Doris Y. Tsao, PhD, challenges previous assumptions derived from coarser brain imaging techniques like fMRI and opens doors to new understandings of memory, creativity, and psychiatric disorders.
The research involved recording activity from 714 individual neurons in the VTC of 16 patients undergoing surgical treatment for epilepsy. These patients, who had electrodes implanted to localize seizure foci, viewed images of common objects such as faces, animals, plants, text, and man-made items. Later, six of them performed an imagination task, mentally visualizing the same objects from memory without visual stimuli. Remarkably, about 40% of the neurons that responded during viewing reactivated with the same pattern during imagination, recreating the precise 'code' used for perception.
Understanding the Ventral Temporal Cortex: The Brain's Visual Library
The ventral temporal cortex, often abbreviated VTC, is a critical hub in the brain's visual processing stream. Located in the inferior temporal lobe, particularly the fusiform gyrus, it specializes in high-level object recognition—distinguishing a face from a car or a tree from a house. Damage here leads to prosopagnosia (face blindness) or other agnosias, underscoring its role in everyday vision.
Previous animal studies, notably by Tsao at Caltech and UC Berkeley, revealed that VTC neurons in macaques use a 'distributed axis code.' In this model, each neuron prefers a specific dimension or 'axis' in an abstract object space—much like coordinates on a map. A population of such neurons collectively represents complex objects through their projections onto these axes. The Cedars-Sinai team confirmed this code operates similarly in humans, with approximately 80% of visually responsive neurons (367 out of 456) exhibiting axis selectivity.
To decode this, researchers leveraged artificial intelligence. Deep neural networks, trained on object classification, generated a low-dimensional 'object space.' Synthetic images maximized along a neuron's preferred axis elicited peak responses, validating the model. This AI-human brain synergy allowed reconstruction of viewed objects from neural activity alone, a feat previously limited to animals.
Methodology: Pioneering Single-Neuron Recordings in Epilepsy Patients
Human single-neuron recordings are rare due to ethical constraints, typically limited to epilepsy monitoring. Patients with drug-resistant epilepsy receive hybrid depth electrodes (microwires) in the VTC for 1-2 weeks to map seizure onset before resection. Rutishauser's team, in collaboration with neurosurgeon Adam Mamelak, MD, capitalized on this window, conducting tasks during downtime.
Step-by-step process:
- Viewing task: Patients saw 192 unique images across categories, repeated in blocks.
- Axis tuning identification: AI analyzed responses to fit the axis model; neurons with significant tuning proceeded.
- Synthetic validation: Generated images probed axis preferences.
- Imagery task: Patients closed eyes, imagined cued objects (e.g., 'apple'), fixating a central point.
- Analysis: Compared imagery responses to viewing; reactivation quantified by projection onto viewing axes.
Results showed imagery responses correlated with viewing axes (r=0.3-0.5), enabling imagery reconstruction distinguishable from random patterns. Mean response amplitudes matched perception, though fewer neurons fired, explaining imagery's fainter quality.
Key Results: 40% Neural Reactivation and Object Reconstruction
Of 456 visually selective neurons, 367 (80%) were axis-tuned. During imagination, 43 of 107 tuned neurons (40%) reactivated significantly. Responses scaled with object-axis alignment, just as in viewing. Using the population code, researchers reconstructed imagined objects, matching category (e.g., face-like patterns for faces).
This reactivation supports a 'generative model' hypothesis: the brain stores abstract representations upstream (e.g., semantics in hippocampus), reactivating VTC sensory codes to generate vivid imagery. Wadia noted, 'We generate a mental image by reactivating the brain cells we used to see it.'
| Metric | Value |
|---|---|
| Total neurons recorded | 714 |
| Visually selective | 456 (64%) |
| Axis-tuned | 367 (80% of selective) |
| Reactivated in imagery | 43/107 (40%) |
Building on Primate Research: From Macaques to Humans
Tsao's prior work identified the axis code in macaque inferotemporal (IT) cortex, akin to human VTC. The human study confirms translational validity—the code is conserved, enabling cross-species models. Rutishauser emphasized, 'It was surprising how well the macaque model mapped to humans.'
Unlike fMRI's population averages, single-neuron data resolves fine-grained reactivation, previously inferred indirectly. Related human studies (e.g., MUSC 2020) showed imagery-perception overlap via AI decoders, but lacked neuron identity.
Clinical Implications: Revolutionizing Treatment for PTSD and OCD
Intrusive imagery haunts PTSD (flashbacks) and OCD (obsessions). Disrupted reactivation may underlie these; therapies could target VTC modulation via neurofeedback or TMS. Mamelak noted potential for 'new therapies for mental conditions involving uncontrolled vivid imagery.'
Aphantasia (2-5% prevalence, inability to visualize) may reflect failed reactivation, informing cognitive training. Memory disorders like Alzheimer's could benefit from boosting reactivation signals.
Read the full study in ScienceAI's Role: Decoding and Generating Neural Visions
Deep networks created object embeddings; generative AI (e.g., VAEs) produced stimuli matching neural preferences. This 'neural reverse engineering' predicts responses to novel images, advancing brain-machine interfaces and prosthetics for blind individuals evoking 'inner vision.'
In education, AI could simulate neural feedback for visual learning, aiding STEM visualization.
Cedars-Sinai's Neuroscience Legacy and Collaborations
Rutishauser's Center for Neural Science and Medicine fosters such interdisciplinary work with Caltech/UCLA. The BRAIN Initiative funded this, highlighting public investment's impact. As an academic health center, Cedars-Sinai trains neurosurgeons and researchers, producing leaders like Wadia (Caltech PhD '23).
Implications for Higher Education and Neuroeducation
Visual imagery underpins learning: mental rotation in math/physics, anatomical recall in medicine. Understanding shared codes suggests imagery training enhances retention—e.g., 'visualize as you study' protocols. Universities could integrate VR for VTC activation, improving outcomes in visual-heavy fields.
Neuroeducation leverages this: active visualization boosts memory consolidation. For faculty, insights inform curriculum design; explore opportunities in neuroscience research at institutions like Cedars-Sinai.
Future Directions: Unlocking the Triggers of Imagination
What cues reactivation? Hippocampal theta rhythms or prefrontal top-down signals? Ongoing studies probe these, potentially via optogenetics analogs or high-density arrays. Longitudinal tracking post-surgery assesses code stability.
Rutishauser: 'This collaboration exemplifies Caltech-Cedars-Sinai synergy.' Expect AI-enhanced therapies and neuroprosthetics.
Photo by Marios Dessign on Unsplash
Why This Matters: A New Era in Cognitive Neuroscience
This study demystifies imagination's biology, affirming the brain as a generative simulator. For higher education, it underscores interdisciplinary research's value, training next-gen neuroscientists. As Wadia reflects, 'The visual system replays states for mental images.' Explore neuroscience careers to contribute.