Fluoroscopy

Fluoroscopy

Join the Competition Contribute to Benchmark

Standard reconstruction benchmark — forward model perfectly known, no calibration needed. Score = 0.5 × clip((PSNR−15)/30, 0, 1) + 0.5 × SSIM

# Method Score PSNR (dB) SSIM Source
🥇 DiffFluoro 0.897 40.0 0.960 ✓ Certified Gao et al. 2024
🥈 PhysFluoro 0.870 38.7 0.949 ✓ Certified Chen et al. 2024
🥉 SwinFluoro 0.847 37.6 0.940 ✓ Certified Li et al. 2023
4 TransFluoro 0.816 36.2 0.925 ✓ Certified Wang et al. 2022
5 REDCNN-Fluoro 0.764 34.0 0.895 ✓ Certified Chen et al. 2017
6 DnCNN-Fluoro 0.718 32.1 0.866 ✓ Certified Chen et al. 2017
7 TV-Fluoro 0.657 29.6 0.828 ✓ Certified Sidky & Pan 2008
8 NLM-Fluoro 0.602 27.4 0.791 ✓ Certified Buades et al. 2005
9 BM3D-Fluoro 0.561 25.8 0.762 ✓ Certified Dabov et al. 2007

Dataset: PWM Benchmark (9 algorithms)

Blind Reconstruction Challenge — forward model has unknown mismatch, must calibrate from data. Score = 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

# Method Overall Score Public
PSNR / SSIM
 Dev
PSNR / SSIM
 Hidden
PSNR / SSIM
 Trust Source
🥇 PhysFluoro + gradient 0.790
0.831
36.33 dB / 0.975
0.782
33.2 dB / 0.954
0.756
30.84 dB / 0.929
✓ Certified Chen et al., IEEE TMI 2024
🥈 TransFluoro + gradient 0.774
0.802
34.27 dB / 0.963
0.764
31.68 dB / 0.939
0.755
31.57 dB / 0.938
✓ Certified Wang et al., IEEE TMI 2022
🥉 DiffFluoro + gradient 0.773
0.847
37.61 dB / 0.981
0.757
30.79 dB / 0.928
0.714
28.39 dB / 0.889
✓ Certified Gao et al., MICCAI 2024
4 SwinFluoro + gradient 0.738
0.816
34.73 dB / 0.966
0.733
30.17 dB / 0.919
0.666
25.97 dB / 0.831
✓ Certified Li et al., Med. Phys. 2023
5 REDCNN-Fluoro + gradient 0.697
0.771
32.02 dB / 0.943
0.684
26.65 dB / 0.849
0.635
25.29 dB / 0.811
✓ Certified Chen et al., IEEE TMI 2017
6 DnCNN-Fluoro + gradient 0.675
0.763
30.4 dB / 0.923
0.657
26.02 dB / 0.833
0.606
24.37 dB / 0.781
✓ Certified Chen et al., IEEE TMI 2017
7 NLM-Fluoro + gradient 0.638
0.650
25.14 dB / 0.807
0.649
25.45 dB / 0.816
0.616
23.69 dB / 0.757
✓ Certified Buades et al., CVPR 2005
8 BM3D-Fluoro + gradient 0.571
0.610
23.4 dB / 0.746
0.572
22.61 dB / 0.715
0.531
21.31 dB / 0.660
✓ Certified Dabov et al., IEEE TIP 2007
9 TV-Fluoro + gradient 0.527
0.694
27.13 dB / 0.861
0.489
19.1 dB / 0.555
0.399
16.35 dB / 0.418
✓ Certified Sidky & Pan, Phys. Med. Biol. 2008

Complete score requires all 3 tiers (Public + Dev + Hidden).

Join the competition →
Scoring: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖) PSNR 40% · SSIM 40% · Consistency 20%
Public 3 scenes

Full-access development tier with all data visible.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

How to use: Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

What to submit: Reconstructed signals (x_hat) and corrected spec as HDF5.

Public Leaderboard
# Method Score PSNR SSIM
1 DiffFluoro + gradient 0.847 37.61 0.981
2 PhysFluoro + gradient 0.831 36.33 0.975
3 SwinFluoro + gradient 0.816 34.73 0.966
4 TransFluoro + gradient 0.802 34.27 0.963
5 REDCNN-Fluoro + gradient 0.771 32.02 0.943
6 DnCNN-Fluoro + gradient 0.763 30.4 0.923
7 TV-Fluoro + gradient 0.694 27.13 0.861
8 NLM-Fluoro + gradient 0.650 25.14 0.807
9 BM3D-Fluoro + gradient 0.610 23.4 0.746
Spec Ranges (3 parameters)
Parameter Min Max Unit
motion_blur -5.0 10.0 ms
lag -3.0 6.0 ms
gain_drift -0.5 1.0 %
View Details →
Dev 3 scenes

Blind evaluation tier — no ground truth available.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), and spec ranges only.

How to use: Apply your pipeline from the Public tier. Use consistency as self-check.

What to submit: Reconstructed signals and corrected spec. Scored server-side.

Dev Leaderboard
# Method Score PSNR SSIM
1 PhysFluoro + gradient 0.782 33.2 0.954
2 TransFluoro + gradient 0.764 31.68 0.939
3 DiffFluoro + gradient 0.757 30.79 0.928
4 SwinFluoro + gradient 0.733 30.17 0.919
5 REDCNN-Fluoro + gradient 0.684 26.65 0.849
6 DnCNN-Fluoro + gradient 0.657 26.02 0.833
7 NLM-Fluoro + gradient 0.649 25.45 0.816
8 BM3D-Fluoro + gradient 0.572 22.61 0.715
9 TV-Fluoro + gradient 0.489 19.1 0.555
Spec Ranges (3 parameters)
Parameter Min Max Unit
motion_blur -6.0 9.0 ms
lag -3.6 5.4 ms
gain_drift -0.6 0.9 %
Submit Reconstruction View Details →
Hidden 3 scenes

Fully blind server-side evaluation — no data download.

What you get & how to use

What you get: No data downloadable. Algorithm runs server-side on hidden measurements.

How to use: Package algorithm as Docker container / Python script. Submit via link.

What to submit: Containerized algorithm accepting y + H, outputting x_hat + corrected spec.

Hidden Leaderboard
# Method Score PSNR SSIM
1 PhysFluoro + gradient 0.756 30.84 0.929
2 TransFluoro + gradient 0.755 31.57 0.938
3 DiffFluoro + gradient 0.714 28.39 0.889
4 SwinFluoro + gradient 0.666 25.97 0.831
5 REDCNN-Fluoro + gradient 0.635 25.29 0.811
6 NLM-Fluoro + gradient 0.616 23.69 0.757
7 DnCNN-Fluoro + gradient 0.606 24.37 0.781
8 BM3D-Fluoro + gradient 0.531 21.31 0.66
9 TV-Fluoro + gradient 0.399 16.35 0.418
Spec Ranges (3 parameters)
Parameter Min Max Unit
motion_blur -3.5 11.5 ms
lag -2.1 6.9 ms
gain_drift -0.35 1.15 %
Submit Algorithm View Details →

Blind Reconstruction Challenge

Challenge

Given measurements with unknown mismatch and spec ranges (not exact params), reconstruct the original signal. A method must be evaluated on all three tiers for a complete score. Scored on a composite metric: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖).

Input

Measurements y, ideal forward model H, spec ranges

Output

Reconstructed signal x̂

About the Imaging Modality

Fluoroscopy provides real-time continuous X-ray imaging for guiding interventional procedures. The forward model is the same Beer-Lambert projection as radiography but at much lower dose per frame (typically 1 uGy/frame at 15-30 fps) resulting in severely photon-limited images. Temporal redundancy from the video stream enables frame-to-frame denoising and recursive filtering. Primary challenges include low SNR, motion blur from patient/organ movement, and veiling glare from scatter.

Principle

Fluoroscopy provides real-time continuous X-ray imaging for guiding interventional procedures. A pulsed or continuous X-ray beam produces live projection images at 7.5-30 fps on a flat-panel detector. The trade-off is between frame rate, radiation dose, and image quality. Temporal filtering and dose-saving modes reduce patient exposure while maintaining diagnostic quality.

How to Build the System

A C-arm fluoroscopy unit has an X-ray tube and flat-panel detector on a C-shaped gantry that can rotate around the patient. Modern systems use pulsed fluoroscopy (variable pulse rate 3.75-30 fps) with automatic brightness control. Install last-image-hold and virtual collimation features. Calibrate geometric distortion for 3-D cone-beam reconstruction capability. Regular dosimetry checks (DAP meter calibration) are mandatory.

Common Reconstruction Algorithms

  • Recursive temporal averaging (IIR filtering for noise reduction)
  • Contrast-enhanced subtraction (road-mapping for angiography)
  • Motion-compensated temporal filtering
  • Cone-beam CT reconstruction from rotational fluoroscopy runs
  • Deep-learning frame interpolation for reduced pulse-rate operation

Common Mistakes

  • Excessive radiation dose from unnecessarily high frame rate or continuous mode
  • Image lag / ghosting from slow detector response at low dose
  • Geometric distortion from C-arm flex not calibrated
  • Scatter degrading contrast in lateral or oblique views of thick anatomy
  • Patient skin dose exceeding threshold (2 Gy) during long procedures

How to Avoid Mistakes

  • Use lowest acceptable pulse rate; employ last-image-hold instead of continuous fluoro
  • Use fast flat-panel detectors (GOS or CsI with fast readout) to minimize lag
  • Perform regular geometric calibration with a phantom for accurate 3D reconstruction
  • Collimate tightly and use appropriate anti-scatter grids
  • Monitor cumulative dose (DAP) and skin dose during procedures; rotate beam angles

Forward-Model Mismatch Cases

  • The widefield fallback applies additive Gaussian blur, but fluoroscopy follows X-ray Beer-Lambert attenuation with real-time temporal dynamics — the exponential transmission model and dynamic contrast are absent
  • Fluoroscopy operates at much lower dose rates than radiography, requiring modeling of quantum mottle (Poisson noise at very low photon counts) and image intensifier/flat-panel detector gain — the widefield noise model is wrong

How to Correct the Mismatch

  • Use the fluoroscopy operator implementing real-time X-ray transmission: y = I_0 * exp(-A*x) with Poisson quantum noise, modeling the low-dose regime and detector response
  • Apply temporal filtering (recursive averaging) or deep-learning denoising tuned for the correct Poisson noise level of fluoroscopic sequences

Experimental Setup — Signal Chain

Experimental setup diagram for Fluoroscopy

Experimental Setup

Instrument: Siemens Artis Pheno / GE Innova IGS 630
Image Size: 1024x1024
Kvp: 70
Frame Rate Fps: 15
Dose Per Frame Ugy: 1.0
Detector Size Cm: 30x30
Detector Type: flat-panel (CsI + aSi)

Key References

  • Defined by IEC 62220-1 standard for fluoroscopy detector characterization

Canonical Datasets

  • Clinical fluoroscopy sequences (institution-specific)

Spec DAG — Forward Model Pipeline

Π(proj) → Σ_t → D(g, η₁)

Π X-ray Projection (proj)
Σ Temporal Integration (t)
D Image Intensifier (g, η₁)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
σ_t motion_blur Temporal motion blur (ms) 0 5.0
τ lag Detector lag time constant (ms) 0 3.0
Δg gain_drift Gain drift per frame (%) 0 0.5

Credits System

40%
Platform Profit Pool
Revenue allocated to benchmark rewards
30%
Winner Share
Top algorithm receives from pool
$100
Min Withdrawal
Minimum payout threshold
Spec Primitives Reference (11 primitives)
P Propagation

Free-space or medium propagation kernel (Fresnel, Rayleigh-Sommerfeld).

M Mask / Modulation

Spatial or spatio-temporal amplitude modulation (coded aperture, SLM pattern).

Π Projection

Geometric projection operator (Radon transform, fan-beam, cone-beam).

F Fourier Sampling

Sampling in the Fourier / k-space domain (MRI, ptychography).

C Convolution

Shift-invariant convolution with a point-spread function (PSF).

Σ Summation / Integration

Summation along a physical dimension (spectral, temporal, angular).

D Detector

Sensor readout with gain g and noise model η (Gaussian, Poisson, mixed).

S Structured Illumination

Patterned illumination (block, Hadamard, random) applied to the scene.

W Wavelength Dispersion

Spectral dispersion element (prism, grating) with shift α and aperture a.

R Rotation / Motion

Sample or gantry rotation (CT, electron tomography).

Λ Wavelength Selection

Spectral filter or monochromator selecting a wavelength band.