NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

China's New Brain Camera Weighs One Gram and Watches 8,500 Neurons at Once

A Tsinghua-led team built a wireless microscope that weighs 1.0 gram without its battery and records thousands of cortical neurons in mice while they climb, swim and fight — a tool for brain science, not a brain implant.

2026-11-19 · 1113 words · NeuroAI
China's New Brain Camera Weighs One Gram and Watches 8,500 Neurons at Once

Two mice meet head-on inside a clear tube. They push, hold, push again. Then one turns and backs out, and the contest is over.

On each animal's head sits a microscope weighing 1.0 gram without its battery. While the mice shove each other, the device is recording thousands of neurons across the cortex. And the neural activity separates winner from loser before the mice have finished deciding — early enough that a model can predict the outcome from brain signals alone.

That experiment is the most memorable result in a paper published in Nature on October 19 by a team at Tsinghua University and the Chinese PLA General Hospital. The device, called Airscope, is a wireless mesoscopic microscope (介观显微镜): small enough to ride on a mouse, wide enough to see a 6-millimetre field, sharp enough to pick out single cells.

It is not a brain-computer interface (脑机接口) in the clinical sense. Nobody is decoding a paralysed patient's intention to move. But the same engineering problem sits underneath both fields: how do you watch a working brain without stopping the behaviour you want to understand?

Key takeaways

  • Airscope weighs 1.0 gram without its battery and records more than 8,500 cortical neurons in a freely moving mouse, roughly 25 times the yield of earlier wireless head-mounted devices.
  • It images a 6 mm field at about 4-micrometre lateral resolution and 12.5 frames per second, close to what benchtop one-photon mesoscopes deliver, but on an animal that is climbing, swimming or socialising.
  • The team used it to predict the winner of a dominance contest from cortical activity, and to record brain and spinal cord simultaneously in a moving animal.
  • This is a laboratory instrument for mice. It is not approved for human use, and none of the paper's claims extend to people.

The trade-off the field has lived with

Neuroscientists have wanted the same thing for two decades: a wide view, single-cell detail, and an animal that behaves normally. Pick two.

Benchtop microscopes give you the first two. The mouse's head is fixed under the objective, or tethered by a cable, and the behaviour you can study shrinks to whatever survives that constraint. Head-mounted miniaturised microscopes give you the third, but typically with a field under 1 millimetre and a weight above 3 grams — enough to change how a mouse moves, grooms and interacts.

Airscope attacks all three at once. The optics use four aspheric lenses to hold a 6 mm field at single-cell resolution, with a total optical track about 65% shorter than earlier miniaturised scopes. A 2-megapixel sensor compresses data on the sensor itself, roughly 25-fold, so a megapixel-scale video stream can travel over a wireless link instead of a cable. The whole unit is more than 30% lighter than the most recent head-mounted mesoscopes, and the magnetic battery can be swapped in about two seconds.

Weight is not a cosmetic spec here. The team reports that mice wearing Airscope showed no significant difference in distance travelled, speed or path curvature compared with mice wearing nothing. That sentence is doing a lot of work: it is the claim that the instrument is not secretly running the experiment.

What they actually did with it

The applications are where the paper earns its place.

  • A room-scale habitat. Mice explored a 5-square-metre enriched arena — ramps, tunnels, shelters — while cortical activity was recorded continuously. Most neural imaging happens in a box the size of a shoebox.
  • A 32-end maze. Long, self-directed exploration, tracked over time rather than in a five-minute trial.
  • Social dominance. Two mice, two scopes, one tube. Dorsal cortical activity predicted the winner of that same contest before the animals had settled it.
  • Brain and spinal cord together. Two devices on one animal allowed the team to characterise signal flow from cord to cortex during fear conditioning.

The paper also reports stable recording from the same cortical region for more than 200 days, which matters for any study that wants to follow learning or disease over months.

Why a microscope is a brain-interface story

It is tempting to file this under basic science and move on. That would miss the point.

Every implanted brain interface is limited by what it can see. Electrodes give you exquisite timing from a small number of neurons. Functional imaging gives you millions of neurons at terrible temporal resolution. A wireless mesoscope that follows a behaving animal sits between those poles, and it is the tool you need to work out which cells carry which intention before you build a decoder.

Dai Qionghai (戴琼海), the Tsinghua academician who led the work, framed it in the university's announcement as a step toward optical brain-computer interfaces with more precise neural detection, and toward immunology and oncology applications where the same device could track disease in a moving body.

Honest limitations

This article is based on the Nature paper's abstract and the published methodology summary, plus Tsinghua's own announcement and Chinese state media coverage of it. I have not read the full methods, the supplementary figures, or the reviewer reports, and I have not seen independent replication.

Some numbers are stated slightly differently by different sources. The paper gives 4-micrometre lateral resolution; Tsinghua's release says 4 to 6 micrometres. The "25-fold" and "30% lighter" comparisons are against specific prior devices chosen by the authors, not against every head-mounted microscope ever built.

The behavioural results are from mice. Mouse cortex is not human cortex, and social dominance in a tube is not a clinical endpoint. Nothing here shortens the path to a human implant, and nothing here has been reviewed by a regulator.

There is also an institutional interest to keep in mind: the university that built the device is also the institution announcing how well it works.

What readers can do now

  • If you follow neurotech, separate tools from therapies. A better microscope improves the science that precedes a device. It does not itself treat anyone, and headlines that blur the two are selling you a story.
  • If you are evaluating "world's first" claims, check the comparison class. Ask which devices were compared, at what weight, with what field of view, and whether the comparison was run by the same lab.
  • If you work in imaging or robotics, look at the compression trick. Doing 25-fold compression on the sensor, before transmission, is the unglamorous decision that made the wireless link possible — and it generalises well beyond neuroscience.

The mice in the tube do not know they are wearing an instrument. They push, one gives way, and somewhere in the cortex the outcome is already written. The interesting part is that we can now watch it happen without holding the animal still.

Related coverage

Cite this article

NeuroAI (脑机网), "China's New Brain Camera Weighs One Gram and Watches 8,500 Neurons at Once," 19 November 2026, https://neuroai.site/a/na-brain-airscope-1gram-wireless-microscope

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Brain–Computer Interface” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories