An ultrafast plenoptic-camera system for high-resolution 3D particle tracking in unsegmented scintillators

This paper presents an ultrafast plenoptic-camera system utilizing time-resolving single-photon avalanche diode arrays to achieve high-resolution 3D particle tracking in large, unsegmented scintillators, offering a cost-effective alternative to traditional dense segmentation for applications ranging from neutrino detection to medical imaging.

The Gist

Imagine you are trying to figure out exactly where a firefly landed inside a giant, dark, foggy room filled with thick jelly. You can't see the firefly directly, but every time it flutters, it flashes a tiny, brief spark of light. Your goal is to map the firefly's path in 3D space just by catching those fleeting sparks.

This is the challenge scientists face when studying neutrinos (ghostly particles that pass through everything) or dark matter. They need to see exactly where these particles interact inside massive blocks of special "jelly" called scintillators.

Here is how the PLATON system described in this paper solves this problem, explained simply:

1. The Problem: The "Too Many Wires" Dilemma

Traditionally, to see inside a giant block of jelly, scientists slice it into millions of tiny cubes (like a 3D grid of sugar cubes). Each cube needs its own wire to report what it sees.

• The Issue: For a detector the size of a swimming pool, you would need millions of wires and electronic boxes. It's incredibly expensive, hard to build, and the wires themselves get in the way.

2. The Solution: The "Magic Eye" Camera

Instead of slicing the jelly, the researchers kept it as one giant, solid block. To see inside it, they invented a new kind of camera system called PLATON.

Think of a standard camera as a single eye. It takes a flat picture. If you want to know how far away something is, you usually need two eyes (stereoscopic vision) or you have to move the camera around.

PLATON uses a "Plenoptic" camera.

• The Analogy: Imagine a standard camera lens is a single window. A Plenoptic camera puts a grid of thousands of tiny, miniature windows (micro-lenses) right in front of the sensor.
• How it works: Each tiny window sees the scene from a slightly different angle, like a crowd of people all looking at a stage from slightly different seats. Even though the camera only takes one photo, the computer can look at all those tiny different angles and mathematically reconstruct the 3D depth. It's like having a "magic eye" that can focus on anything in the room after the photo is taken.

3. The Super-Sensor: The "Photon Hunter"

The light from these particles is incredibly faint—sometimes just a few sparks (photons) are emitted. A normal camera would see nothing but blackness.

• The Tech: PLATON uses a SPAD array. Think of this not as a camera sensor, but as a massive grid of millions of individual, ultra-sensitive "ears."
• The Speed: These ears can hear a single "click" (a photon) and tell you exactly when it happened, down to a trillionth of a second. This speed is crucial because it helps filter out the "static" (noise) and keeps only the real signals.

4. The Process: Reconstructing the Ghost

Here is how they track a particle:

1. The Flash: A neutrino zips through the giant jelly block and hits an atom, creating a tiny burst of blue light.
2. The Capture: The PLATON cameras, sitting outside the block, catch these faint sparks through their grid of micro-lenses.
3. The Ray Tracing: The computer acts like a detective. It takes every single spark it caught and draws a line backward through the micro-lenses to guess where it came from.
4. The 3D Map: By combining thousands of these backward lines, the computer finds where they all cross. That crossing point is exactly where the particle was!

5. The "AI Brain"

Because there are so many sparks and so much data, a human couldn't sort it out. The team used a Deep Learning AI (specifically a "Transformer," the same type of brain used in advanced chatbots).

• The Analogy: Imagine trying to find a specific path in a forest by looking at a million scattered leaves. A human would get lost. The AI looks at the pattern of the leaves and instantly says, "Ah, these leaves were kicked up by a deer running this way."
• The AI learned to connect the dots between the faint sparks to draw the particle's path with incredible precision (about the width of a human hair).

Why This Matters

• No More Slicing: You can build massive detectors without cutting them up, saving money and complexity.
• Super Sharp Vision: They achieved a resolution of 200 micrometers (0.2 mm). That's like seeing a grain of sand from a distance of 20 meters.
• Future Applications: This isn't just for physics. This technology could revolutionize medical imaging (like better CT scans), finding hidden objects (neutron radiography), and even looking for dark matter.

In a nutshell: The researchers replaced a messy, expensive grid of wires with a single, super-fast, "magic eye" camera system powered by AI. This allows them to see the invisible paths of ghost particles inside giant blocks of material with stunning clarity.

Technical Summary

1. Problem Statement

Modern high-energy physics experiments (e.g., neutrino detectors, dark matter searches, calorimeters) require massive, dense active volumes to detect weakly interacting particles. However, achieving high-precision 3D particle tracking (spatial resolution $\sim$100 $\mu$m) in these large volumes presents significant challenges:

• Segmentation Costs: Traditional approaches rely on fine geometric segmentation (e.g., scintillator fibers or blocks) coupled with thousands of readout channels. This leads to prohibitive construction complexity and costs, especially for tonne-scale detectors.
• Readout Bottlenecks: Dense segmentation requires a vast number of analog readout channels, each needing independent digitization electronics.
• Limitations of Existing Optical Methods: Previous attempts to image particle events in unsegmented volumes using conventional cameras or Monte Carlo simulations have struggled with limited depth resolution, inability to reconstruct tracks outside a single lens's field of view, or lack of single-photon sensitivity and timing.

The core challenge is to achieve sub-millimeter 3D spatial resolution and sub-nanosecond timing in a monolithic (unsegmented) scintillator volume without the cost and complexity of massive channel counts.

2. Methodology

The authors propose a paradigm shift using a PLATON (PLenoptic imAge of Tracked photONs) detector concept. This system combines two enabling technologies:

1. Plenoptic Cameras (Light Field Cameras): These use a main objective lens and a Micro-Lens Array (MLA) placed in front of the sensor. Each micro-lens captures a slightly different perspective of the scene, encoding both spatial ($x, y, z$) and angular ($\phi, \theta$) information of light rays in a single exposure. This allows for depth reconstruction from a single 2D image.
2. SPAD Array Sensors: Single-Photon Avalanche Diode (SPAD) arrays provide single-photon sensitivity with sub-nanosecond time resolution (timestamping). Unlike traditional Silicon Photomultipliers (SiPMs), SPAD arrays integrate readout electronics on-chip, allowing millions of pixels to be read out via a few data lines, drastically reducing channel count.

Key Technical Components:

• Prototype (PLATON-prototype): Utilized a SwissSPAD2 sensor (16.39 $\mu$m pitch, 5% PDE) and a Raytrix MLA (f/2.4).
• Post-Processing: Developed a ray-tracing algorithm that backtracks detected photons from the sensor pixels through the micro-lens centers to the object space. Outliers (noise) are iteratively rejected to triangulate the photon origin.
• Deep Learning Reconstruction: For complex multi-particle events (neutrino interactions), the authors developed a Transformer-based Neural Network. This architecture treats detected photons as tokens, using attention mechanisms to capture global spatial correlations and hyper-dimensional relationships between sparse photon hits, outperforming traditional CNNs or GNNs for this specific sparse data topology.
• Simulation: Extensive Geant4 simulations were performed for a 10 cm module (PLATON-10cm) and a 1 m$^3$ scale detector (PLATON-1m), modeling organic scintillator (EJ-262) properties and neutrino interactions (T2K beam flux).

3. Key Contributions

• First Demonstration of Plenoptic Imaging for Particle Tracking: The paper presents the first experimental demonstration of using a plenoptic camera with a SPAD array to reconstruct the 3D origin of single photons and particle tracks in a scintillator.
• Photon-Starved Imaging: Developed robust post-processing methods capable of reconstructing 3D positions from extremely low photon counts (photon-starved images), a critical requirement for scintillation light detection.
• Deep Learning for Sparse Photon Data: Introduced a Transformer-based architecture specifically adapted for reconstructing 3D particle tracks from sparse, unordered photon detection events, achieving high precision where traditional methods fail.
• Scalable Detector Concept: Provided a design recipe and simulation validation for scaling the concept from a prototype to a 1-tonne scale detector, demonstrating that high resolution is achievable without geometric segmentation.

4. Results

Prototype Performance (PLATON-prototype):

• Spatial Resolution: Achieved a 3D spatial resolution of 2.4 mm (calibration sample) and 4.5 mm (validation sample) for point-like light sources.
• Photon Dependence: With 10 detected photons, the 2D lateral resolution dropped below 1.2 mm.
• Event Reconstruction: Successfully reconstructed the position of $^{90}$Sr electrons in a plastic scintillator block. The reconstructed positions were within 20 mm of the true source, limited primarily by depth resolution and the out-of-focus setup.

Simulation Results (PLATON-10cm & PLATON-1m):

• Neutrino Tracking: In a simulated 10 cm module exposed to GeV muon neutrinos, the system achieved an average 3D tracking resolution of ~190 $\mu$m (ranging 150–340 $\mu$m).
• Vertex Resolution: The interaction vertex resolution for charged-current (CC) interactions was 0.42 mm.
• Particle Identification: The system could distinguish between muons and protons with high efficiency. For CC $1\mu0\pi2p$ events (a "golden channel" for nuclear physics), the selection purity was 90% up to 1.4 GeV/c.
• Comparison with Classical Cameras: The plenoptic system significantly outperformed an equivalent array of classical cameras. For a 1 m$^3$ volume, the plenoptic system offered ~4x better spatial resolution than classical cameras. Classical cameras failed to reconstruct ~23% of events when the source was outside a single lens's field of view, whereas plenoptic cameras succeeded in all events.
• Scalability: Simulations for a 1 m$^3$ detector showed a resolution of 3.7 mm for 1 MeV energy loss, improving to 1.5 mm for 10 MeV. The authors note that with optimized optics and smaller pixels, sub-millimeter resolution is achievable at this scale.

5. Significance and Future Outlook

This work establishes a viable alternative to traditional segmented detectors for high-energy physics.

• Cost and Complexity Reduction: By replacing thousands of readout channels with a few plenoptic cameras and SPAD arrays, the construction cost and complexity for tonne-scale detectors are drastically reduced.
• New Physics Capabilities: The high spatial and temporal resolution enables precise measurement of neutrino-nucleus cross-sections and the study of complex nuclear effects (e.g., 2p2h interactions) that are currently major sources of systematic uncertainty in neutrino oscillation experiments (like T2K and Hyper-Kamiokande).
• Broader Applications: Beyond neutrino physics, the technology is applicable to:
  • Medical Imaging: Positron Emission Tomography (PET) and proton CT.
  • Neutron Detection: Fast neutron time-of-flight measurements.
  • Dark Matter: Enhanced sensitivity for dark matter candidates.
  • Cherenkov Radiation: Improved reconstruction of light rings.

The paper concludes that the PLATON concept, driven by advances in SPAD array sensors and deep learning, paves the way for a new generation of high-resolution, unsegmented particle detectors capable of operating at the kilotonne scale without the cryogenic infrastructure required by Liquid Argon Time Projection Chambers (LArTPCs).