Xaq Pitkow stands in front of a whiteboard with writing on it. He holds a copy of Nature dedicated to the MICrONS project, and he smiles at the camera.

A Tiny Brain Map is Opening New Opportunities in Neuroscience

By Kirsten Heuring Email Kirsten Heuring

Heidi Opdyke

A piece of brain no larger than a poppy seed is opening a new frontier in neuroscience. A Carnegie Mellon University researcher whose work uses artificial intelligence to connect brain structure, neural activity and behavior played a significant role in the effort.

The research sits at the intersection of neuroscience and artificial intelligence, an emerging field known as NeuroAI, which uses insights from the brain to improve AI systems and uses AI tools to better understand how the brain works. By combining detailed measurements of neural circuits with advanced computational models, researchers hope to uncover the principles that underlie perception, learning and behavior.

Capturing the brain in action

As part of the Machine Intelligence from Cortical Networks (MICrONS) project, a major collaboration co funded by IARPA and the NIH Brain Initiative, scientists mapped a cubic millimeter containing about 200,000 cells and more than 500 million connections. The project is one of the most detailed reconstructions of neural circuitry to date and offers an unprecedented window into how the brain processes visual information. The work could advance understanding of the neural mechanisms behind perception while also informing new approaches to machine learning and artificial intelligence.

“Fifty years ago, people thought this was impossible,” said Xaq Pitkow, professor in Carnegie Mellon’s Neuroscience Institute and a project contributor. “It’s really cool to see that the impossible has manifested.”

The MICrONS project focused on a portion of a mouse’s visual cortex, combining advanced imaging techniques with computational modeling to link brain structure with function. Researchers first recorded how neurons fired in response to video stimuli in real time, then reconstructed the connections among those same neurons to understand how signals flow through the network.

Pitkow's role was to help bridge those different layers of information. His team develops AI-based models that connect the brain's wiring, the activity of individual neurons and the resulting computations that enable perception and behavior. By comparing model predictions with recordings from real neurons, researchers can test ideas about how the brain processes visual information.

Andreas Tolias, a professor in Stanford University’s Department of Ophthalmology and a lead collaborator on the project, used a specialized imaging system known as a two-photon random access mesoscope (2P-RAM) developed at Janelia Research Campus. The instrument allowed researchers to observe large networks of neurons while maintaining the resolution to track individual cells. Tolias’ lab also used new approaches to label the blood vessels, making the resulting recordings more detailed than ever before.

The imaging process was extremely challenging, Tolias said.

“Miniscule changes in room temperature could cause catastrophic levels of microscope drift,” Tolias said. “We had to run this pipeline continuously for weeks, with shift changes to keep it running. The imaging personnel had a baton they would pass to each other on shift changes that said, ‘Always be scanning.’”

From images to intelligence

After the imaging team completed scanning the cortex, the Allen Institute for Brain Science and a team led by Sebastian Seung, a professor at Princeton University, imaged with electron microscopy and reconstructed all the fine anatomical details. Pitkow, Tolias and other modeling researchers then used the resulting data to build computational models that mimic brain function.

Using deep learning and artificial intelligence, they trained models to predict how neurons would respond to visual scenes and compared those predictions with real neural activity.

“You’re trying to predict what the neurons would say in response to video,” Pitkow said. “These mathematical models basically take inputs of video and produce outputs that look like the neurons.”

The MICrONS team published its initial findings in April 2025. Besides sharing their methods and results, they made their computational and anatomical data available to researchers and the public.

“Questions that used to require laborious, bespoke experiments can now be answered by mining this dataset,” Tolias said. “The community will come up with many more novel questions and approaches than any single working group could.”

And members of the community, including Pitkow, are using the data in new ways.

New clues to how we see the world

For Pitkow, the dataset provides a foundation for understanding how the brain balances consistency and context when interpreting visual information.

Scientists have long known that some neurons respond reliably to basic visual features such as the orientation of an edge or line. Other neurons respond more flexibly, taking into account the broader visual scene and more complex patterns.

Using MICrONS data, Pitkow identified neurons that appear to combine both properties. Instead of responding only to a simple feature, these neurons seem to recognize a specific visual pattern while remaining flexible about some of the details surrounding it. They can respond to a consistent feature while also incorporating information about the larger image.

“It’s informative about object boundaries,” Pitkow said.

The dataset also allows researchers to test how neural responses change across different contexts. By presenting different images to computational models, Pitkow can measure how activity shifts from one visual environment to another. Using both simple artificial images, such as lines of light, and more naturalistic images that resemble everyday visual experiences, he found that neurons responded differently depending on what they were viewing.

“How natural the image is really changes the selectivity and the context dependence of these neurons,” Pitkow said.

Pitkow plans to publish additional findings related to MICrONS data later this year. He said the work is only the beginning of what MICrONS could enable.

“By combining these incredibly comprehensive measurements of the brain's structure and function, understanding some fundamental mechanisms of thought may now be within,” Pitkow said.

Pitkow will discuss that future as a featured speaker at the NIH's 2026 BRAIN Initiative Conference, Inventing the Future, held Aug. 11-13, 2026, in Rockville, Maryland. He will participate in the NeuroAI Innovation Domain session, “The BRAIN NeuroAI Roadmap: Closing the Loop Between Natural and Artificial Intelligence.”