Researchers build AI camera that reconstructs particle paths using handful of photons | Technology News
4 min readJul 18, 2026 09:39 PM Researchers at ETH Zurich and EPFL have developed a new particle detection system that could transform how scients observe some of the universe’s most elusive particles. The technology, called PLATON, replaces millions of tiny detector components with a single block of light-producing material and uses an AI-powered camera system to reconstruct particle paths in three dimensions.
The findings, recently published in Nature Communications, suggest the detector could match or even outperform today’s most advanced particle detectors while being significantly easier to build and scale. Researchers also believe the technology could eventually improve medical imaging systems such as PET scanners.
New approach to particle detection
Particle detectors play a crucial role in physics experiments tracking the movement of elementary particles through dense materials. Conventional detectors rely on scintillators, which emit tiny flashes of light when struck charged particles. To pinpoint where those interactions occur, the scintillator is typically divided into millions of small segments connected optical fibres and photon sensors.
While highly accurate, this design becomes increasingly complex and expensive as detectors grow larger. Exing experiments, such as Japan’s T2K neutrino experiment and CERN’s LHCb, use millions of individual components to achieve high spatial resolution.
PLATON takes a different approach using a single, unsegmented scintillator block and reconstructing the origin of the emitted light with advanced camera technology.
Light-field camera combined with AI
The detector is based on a plenoptic, or light-field, camera capable of recording not only the intensity of incoming light but also the direction from which it arrives. Combined with a single-photon avalanche diode (SPAD) sensor, the system can detect individual photons generated inside the scintillator.
Researchers built the proof-of-concept detector using a micro-lens array mounted directly onto the SwissSPAD2 imaging sensor, allowing it to capture extremely faint flashes of light while filtering out background noise through gated photon detection.Story continues below this ad
Laboratory tests showed the prototype could reconstruct particle interactions using as few as five detected photons. Simulations closely matched the experimental results, validating the detector’s performance.
Neural network reconstructs particle paths
The team also developed a neural network based on Transformer architecture, the same family of AI models widely used in large language models, to analyse where and when photons appear inside the detector.
Instead of processing text, the AI identifies correlations among detected photons to reconstruct the original particle interaction. Simulations indicate that a future version of PLATON could achieve sub-millimetre spatial resolution inside a 10×10×10 cm detector while efficiently identifying neutrino interactions.
Researchers also modelled a much larger detector measuring one cubic metre and found it could still achieve a resolution of only a few millimetres, comparable to today’s state-of-the-art scintillator detectors, despite eliminating millions of individual components.Story continues below this ad
Potential uses beyond particle physics
Beyond fundamental physics experiments, the researchers believe PLATON could benefit medical imaging.
The team has already filed three patents covering the use of the technology in positron emission tomography (PET), where more accurate reconstruction of faint light signals could produce sharper medical scans. According to the researchers, PLATON could become another example of particle physics research leading to technologies with broader scientific and healthcare applications.
