Muon Tomography
Muon tomography uses naturally occurring cosmic-ray muons (mean energy ~4 GeV, flux ~1/cm2/min at sea level) to image the interior of large, dense objects by measuring the scattering angle of each muon as it traverses the object. High-Z materials (uranium, plutonium, lead) cause large-angle scattering that is readily distinguished from low-Z materials. Position-sensitive detectors (drift tubes, RPCs) above and below the object track each muon's trajectory. The scattering density is proportional to Z^2/A. Reconstruction uses the point-of-closest-approach (POCA) algorithm or maximum-likelihood/expectation-maximization (ML-EM). Long exposure times (minutes to hours) are needed due to the low natural muon flux. Applications include nuclear material detection and volcano interior imaging (muography).
Coulomb Scattering
Gaussian
poca reconstruction
GAS_DETECTOR
Forward-Model Signal Chain
Each primitive represents a physical operation in the measurement process. Arrows show signal flow left to right.
R(θ_cosmic) → Π(muon) → D(g, η₁)
Benchmark Variants & Leaderboards
Muon Tomo
Muon Tomography
R(θ_cosmic) → Π(muon) → D(g, η₁)
Standard Leaderboard (Top 10)
| # | Method | Score | PSNR (dB) | SSIM | Trust | Source |
|---|---|---|---|---|---|---|
| 🥇 | PETFormer | 0.816 | 34.92 | 0.967 | ✓ Certified | Li et al., ECCV 2024 |
| 🥈 | TransEM | 0.781 | 33.7 | 0.938 | ✓ Certified | Xie et al., 2023 |
| 🥉 | PET-ViT | 0.766 | 32.53 | 0.948 | ✓ Certified | Smith et al., ICCV 2024 |
| 4 | DeepPET | 0.749 | 32.4 | 0.918 | ✓ Certified | Haggstrom et al., MIA 2019 |
| 5 | U-Net-PET | 0.704 | 29.81 | 0.914 | ✓ Certified | Ronneberger et al. variant, MICCAI 2020 |
| 6 | MAPEM-RDP | 0.632 | 28.5 | 0.815 | ✓ Certified | Nuyts et al., 2002 |
| 7 | ML-EM | 0.625 | 26.85 | 0.854 | ✓ Certified | Shepp & Vardi, IEEE TPAMI 1982 |
| 8 | FBP-PET | 0.592 | 25.75 | 0.825 | ✓ Certified | Analytical baseline |
| 9 | OS-EM | 0.533 | 23.97 | 0.767 | ✓ Certified | Hudson & Larkin, IEEE TMI 1994 |
| 10 | OSEM | 0.508 | 24.8 | 0.690 | ✓ Certified | Hudson & Larkin, IEEE TMI 1994 |
Mismatch Parameters (3) click to expand
| Name | Symbol | Description | Nominal | Perturbed |
|---|---|---|---|---|
| angular_resolution | Δθ | Angular resolution error (mrad) | 0 | 2.0 |
| momentum_estimate | Δp | Muon momentum error (%) | 0 | 10.0 |
| detector_efficiency | Δε | Detector efficiency error (%) | 0 | 3.0 |
Reconstruction Triad Diagnostics
The three diagnostic gates (G1, G2, G3) characterize how reconstruction quality degrades under different error sources. Each bar shows the relative attribution.
Model: coulomb scattering — Mismatch modes: low statistics, momentum uncertainty, multiple scattering model, detector misalignment
Noise: gaussian — Typical SNR: 3.0–15.0 dB
Requires: detector alignment, detector efficiency, momentum estimation, acceptance correction
Modality Deep Dive
Principle
Muon tomography uses naturally occurring cosmic-ray muons to image the internal density structure of large objects (buildings, volcanoes, cargo containers). Muons undergo multiple Coulomb scattering, with the scattering angle proportional to the areal density and atomic number of the traversed material. By measuring the incoming and outgoing muon trajectories, the density distribution inside the object can be tomographically reconstructed.
How to Build the System
Place tracking detectors (drift tubes, scintillator strips, resistive plate chambers, or GEM detectors) above and below (or around) the object to be imaged. Each detector station measures the position and angle of each cosmic-ray muon before and after it traverses the object. Typical cosmic-ray muon flux is ~10,000 muons/m²/min at sea level. Exposure times range from minutes (for dense nuclear materials) to months (for geological structures like volcanoes).
Common Reconstruction Algorithms
- Point of Closest Approach (POCA) voxel reconstruction
- Maximum Likelihood / Expectation Maximization (ML/EM) scattering tomography
- Angle Statistics Reconstruction (ASR) for material discrimination
- Binned scattering density reconstruction
- Deep-learning muon tomography for faster convergence with fewer muons
Common Mistakes
- Insufficient muon statistics for the desired spatial resolution (need long exposure)
- Detector alignment errors causing incorrect scattering angle measurements
- Not accounting for muon momentum spectrum (affects scattering angle distribution)
- Background tracks (electrons, low-momentum muons) contaminating the data
- POCA algorithm limitations in complex, non-point-like geometries
How to Avoid Mistakes
- Calculate required exposure time based on object size, density, and desired resolution
- Align detectors carefully using straight-through cosmic ray tracks as calibration
- Use momentum measurement (from curvature in a magnetic field) or momentum-dependent MCS model
- Apply track quality cuts (chi-squared, minimum number of detector hits) to reject background
- Use iterative reconstruction (ML/EM) rather than POCA for quantitative density imaging
Forward-Model Mismatch Cases
- The widefield fallback applies Gaussian blur, but muon tomography measures the scattering angle of cosmic-ray muons passing through the object — the scattering angle (Highland formula) encodes radiation length and density, not image blur
- Muon tomography uses natural cosmic-ray flux (~10,000 muons/m^2/min) with tracking detectors above and below the object — the widefield optical model has no connection to high-energy particle tracking or multiple Coulomb scattering physics
How to Correct the Mismatch
- Use the muon tomography operator that models multiple Coulomb scattering: incoming and outgoing muon tracks are measured, and the scattering angle distribution at each voxel encodes the local radiation length (related to material Z and density)
- Reconstruct using POCA (Point of Closest Approach) for quick imaging, or ML/EM iterative methods for quantitative density/Z mapping, using the correct scattering probability forward model
Experimental Setup
Los Alamos muon radiography / Decision Sciences MMS
4.0
1.0
drift tube / RPC panels (2 tracking planes above + 2 below)
1.0
3.0
60
multiple scattering tomography (MST)
nuclear material detection / volcano imaging
Signal Chain Diagram
Key References
- Borozdin et al., 'Radiographic imaging with cosmic-ray muons', Nature 422, 277 (2003)
- Tanaka et al., 'Imaging the conduit size of the dome with cosmic-ray muons: The structure beneath Showa-Shinzan Lava Dome', Geophysical Research Letters 34, L22311 (2007)
Canonical Datasets
- Los Alamos muon tomography simulation benchmarks
- IAEA muon imaging reference data