Researchers at the University of California, Davis, have developed a miniaturized microscope that enables real-time, high-resolution, and non-invasive imaging of brain activity in mice. The innovation marks a major advancement in how scientists can study the brain in action.
The DeepInMiniMicroscope developed by UC Davis electrical engineering professor Weijian Yang combines optical technology and machine learning to create a device that can take high-resolution three-dimensional images inside living tissue. Image Credit: Mario Rodriguez/UC Davis
What we are doing is creating technology to image brain activity in freely moving and behaving mice to open up the behavior paradigm. The goal is to create a device capable of enabling research into brain activity and behavior in mice in real time — to see how brain activity drives behavior or perception.
Weijian Yang, Professor, Electrical and Computer Engineering, University of California, Davis
This compact device opens new possibilities for exploring the connection between brain function and behavior, with the potential to inform future treatments for neurological and psychiatric disorders.
The microscope system, named DeepInMiniscope, is detailed in a study published in Science Advances.
Iterative Design
DeepInMiniscope builds on Yang’s earlier work, developing a lensless camera that could generate 3D images from a single exposure. That system worked well for applications like robotic vision, where objects are large and lighting conditions are controlled. But when it came to biological samples, particularly in live tissue, the system faced hurdles, including intense light scattering and low image contrast.
To address this, the team created a new optical mask featuring more than 100 tiny, high-resolution lenslets. These work in tandem with a custom-designed neural network that stitches together data from each lenslet to form detailed 3D reconstructions.
Deep (Learning) Insights
The DeepInMiniscope’s neural network is a hybrid model that blends multiple machine learning strategies into what’s known as an “unrolled neural network.” This architecture enables fast, accurate 3D reconstruction of complex brain activity in real time.
Our algorithm combines interpretability, efficiency, scalability and precision. It requires only a minimal amount of training data, yet it can robustly and accurately process large-scale datasets at high speed.
Feng Tian, Study First Author and Postdoctoral Researcher, University of California, Davis
Using this system, the team has successfully captured real-time neuronal activity in freely moving mice; a major milestone for behavioral neuroscience.
Hat Trick
Yang’s vision is a tool small and light enough for mice to wear comfortably as they move naturally. The current prototype measures just 3 square centimeters (roughly the size of a grape) and weighs only 10 grams, about as much as four pennies.
Previous designs relied on bulky, traditional cameras that limited mobility. In contrast, DeepInMiniscope uses a minimalist setup built around a bare image sensor circuit board, significantly reducing the device’s footprint.
The team is already working toward the next version: a wireless, 2-square-centimeter model — essentially a “hat” for a mouse.
“By enabling real-time observation of brain activity in freely behaving mice, this technology not only advances our fundamental understanding of how the brain processes information and drives behavior, but also contributes to improving our understanding of brain disorders and the development of future therapeutic strategies in humans,” concluded Yang.
Journal Reference:
Tian, F., et al. (2025) DeepInMiniscope: Deep learning–powered physics-informed integrated miniscope. Science Advances. doi.org/10.1126/sciadv.adr6687