Scientists Develop AI-Powered Camera That Reconstructs Particle Paths Using Just a Few Photons

Researchers from ETH Zurich and EPFL have unveiled an innovative particle detection system that combines artificial intelligence with advanced imaging technology to reconstruct particle trajectories using only a handful of photons. The breakthrough, known as PLATON, could simplify the design of next-generation particle detectors while opening new possibilities for medical imaging technologies such as PET scanners.

The research, published in Nature Communications, introduces a radically different approach to particle detection. Instead of relying on millions of tiny detector elements, PLATON uses a single block of scintillating material and an AI-powered camera capable of determining where particle interactions occur in three dimensions.

A Simpler Way to Detect Particles

Traditional particle detectors use scintillators—materials that emit tiny flashes of light when struck by charged particles. To accurately identify where those interactions take place, existing systems divide the scintillator into millions of small segments connected to optical fibres and photon sensors.

While highly precise, these detectors are expensive, difficult to manufacture, and increasingly complex as they scale up. Large scientific experiments, including CERN’s LHCb and Japan’s T2K neutrino experiment, depend on millions of detector components to achieve high-resolution measurements.

PLATON eliminates much of this complexity by replacing segmented detectors with a single scintillator block while relying on sophisticated imaging and AI algorithms to determine the exact origin of emitted light.

AI-Powered Light-Field Camera

At the heart of the system is a plenoptic (light-field) camera, capable of recording not only the brightness of incoming light but also the direction from which each photon arrives.

The researchers paired the camera with a Single-Photon Avalanche Diode (SPAD) imaging sensor, allowing the detector to capture extremely faint flashes produced inside the scintillator. A specially designed micro-lens array mounted directly on the SwissSPAD2 sensor improves photon collection while reducing background noise.

During laboratory testing, the prototype successfully reconstructed particle interactions using as few as five detected photons, with experimental results closely matching computer simulations.

Transformer AI Reconstructs Particle Trajectories

To process the captured photon data, the research team developed a neural network based on the Transformer architecture, the same AI framework widely used in modern large language models.

Rather than analysing text, the AI identifies spatial and temporal relationships between detected photons to reconstruct the original particle’s trajectory.

Simulation results suggest that future versions of PLATON could achieve sub-millimetre spatial resolution inside a detector measuring 10 × 10 × 10 centimetres, while larger one-cubic-metre detectors could still maintain resolution within only a few millimetres—comparable to today’s most advanced scintillator-based detection systems.

Potential Applications in Healthcare

Beyond particle physics, the researchers believe PLATON could significantly improve Positron Emission Tomography (PET) scanners by reconstructing faint light signals more accurately, potentially producing sharper and more precise medical images.

The team has already filed three patents covering medical applications of the technology, highlighting its potential to move beyond laboratory research into real-world healthcare systems.

If successfully commercialized, PLATON could become another example of cutting-edge particle physics research driving innovations that benefit both science and medicine.