Neurons for Seeing and Imagining

Why It Matters

Showing that perception and mental imagery rely on a common axis code links basic neuroscience to clinical strategies for intrusive‑image disorders and fuels next‑generation brain‑computer interfaces.

Key Takeaways

• 714 VTC neurons recorded from 16 epilepsy patients
• 80% of category‑selective neurons encode low‑dimensional axes
• Axis code reconstructs viewed objects and predicts superstimuli responses
• 40% of axis‑tuned neurons reactivate during visual imagination
• Shared neural code may guide treatments for intrusive imagery disorders

Pulse Analysis

Mental imagery—visualizing scenes without external input—has long been recognized as a cornerstone of creativity, memory consolidation, and even problem solving. While functional imaging has mapped broad networks involved in imagination, the precise firing patterns of individual neurons remained elusive. By targeting the ventral temporal cortex, a hub for object recognition, researchers filled this gap, leveraging a rare clinical setting where epilepsy patients consent to intracranial recordings. This approach bridges the macro‑scale insights of fMRI with the micro‑scale precision needed to decode the brain's representational language.

The team recorded from 714 neurons while participants viewed five object categories, identifying 456 cells that responded selectively. Remarkably, 80% of these cells aligned with a low‑dimensional "axis" framework derived from deep neural networks, effectively translating complex visual features into a compact code. Using this axis code, the scientists could reconstruct the visual stimulus from neural activity and even forecast responses to artificially generated superstimuli—images engineered to maximally activate specific neurons. When the same participants imagined the objects, about 40% of the axis‑tuned cells reignited, and their firing strength mirrored the imagined object's projection onto each cell's preferred axis, confirming a shared representational scheme.

These findings reshape our understanding of how perception and imagination intertwine at the cellular level, suggesting that the brain reuses a common coding strategy rather than deploying separate circuits. Clinically, this insight could inform interventions for conditions like PTSD or schizophrenia, where intrusive mental images are debilitating. Moreover, the axis‑based representation offers a promising blueprint for brain‑computer interfaces, enabling more accurate decoding of intended visual content from neural signals. As artificial intelligence continues to model visual processing, the convergence of deep‑network axes and human neuronal codes may accelerate both neuroscience and machine‑vision technologies.