X-ray Radiography

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
🥇	Score-CT Score-CT Song et al., NeurIPS 2024 39.92 dB SSIM 0.984 Checkpoint unavailable	0.907	39.92	0.984	✓ Certified	Song et al., NeurIPS 2024
🥈	DiffusionCT DiffusionCT Kazemi et al., ECCV 2024 39.68 dB SSIM 0.982 Checkpoint unavailable	0.902	39.68	0.982	✓ Certified	Kazemi et al., ECCV 2024
🥉	CTFormer CTFormer Li et al., ICCV 2024 39.45 dB SSIM 0.980 Checkpoint unavailable	0.897	39.45	0.980	✓ Certified	Li et al., ICCV 2024
4	CT-ViT CT-ViT Guo et al., NeurIPS 2024 39.15 dB SSIM 0.978 Checkpoint unavailable	0.891	39.15	0.978	✓ Certified	Guo et al., NeurIPS 2024
5	DOLCE DOLCE Liu et al., ICCV 2023 38.32 dB SSIM 0.971 Checkpoint unavailable	0.874	38.32	0.971	✓ Certified	Liu et al., ICCV 2023
6	DuDoTrans DuDoTrans Wang et al., MLMIR 2022 37.68 dB SSIM 0.962 Checkpoint unavailable	0.859	37.68	0.962	✓ Certified	Wang et al., MLMIR 2022
7	Learned Primal-Dual Learned Primal-Dual Adler & Oktem, IEEE TMI 2018 36.42 dB SSIM 0.947 Checkpoint unavailable	0.831	36.42	0.947	✓ Certified	Adler & Oktem, IEEE TMI 2018
8	FBPConvNet FBPConvNet Jin et al., IEEE TIP 2017 35.81 dB SSIM 0.939 Checkpoint unavailable	0.816	35.81	0.939	✓ Certified	Jin et al., IEEE TIP 2017
9	RED-CNN RED-CNN Chen et al., IEEE TMI 2017 33.56 dB SSIM 0.908 Checkpoint unavailable	0.763	33.56	0.908	✓ Certified	Chen et al., IEEE TMI 2017
10	PnP-DnCNN PnP-DnCNN Zhang et al., 2017 33.45 dB SSIM 0.905 Try in SpecLab →	0.760	33.45	0.905	✓ Certified	Zhang et al., 2017
11	PnP-ADMM PnP-ADMM Venkatakrishnan et al., 2013 32.64 dB SSIM 0.891 Try in SpecLab →	0.740	32.64	0.891	✓ Certified	Venkatakrishnan et al., 2013
12	TV-ADMM TV-ADMM Sidky et al., 2008 30.15 dB SSIM 0.862 Try in SpecLab →	0.683	30.15	0.862	✓ Certified	Sidky et al., 2008
13	FBP FBP Kak & Slaney, 1988 27.38 dB SSIM 0.790 Try in SpecLab →	0.601	27.38	0.790	✓ Certified	Kak & Slaney, 1988

Dataset: PWM Benchmark (13 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
🥇	CTFormer + gradient CTFormer + gradient Li et al., ICCV 2024 Score 0.803 Correct & Reconstruct →	0.803	0.840 36.9 dB / 0.978	0.800 35.44 dB / 0.970	0.770 33.68 dB / 0.958	✓ Certified	Li et al., ICCV 2024
🥈	CT-ViT + gradient CT-ViT + gradient Guo et al., NeurIPS 2024 Score 0.797 Correct & Reconstruct →	0.797	0.836 36.69 dB / 0.977	0.789 34.17 dB / 0.962	0.767 32.42 dB / 0.947	✓ Certified	Guo et al., NeurIPS 2024
🥉	DOLCE + gradient DOLCE + gradient Liu et al., ICCV 2023 Score 0.771 Correct & Reconstruct →	0.771	0.846 36.69 dB / 0.977	0.755 30.79 dB / 0.928	0.711 28.72 dB / 0.895	✓ Certified	Liu et al., ICCV 2023
4	Score-CT + gradient Score-CT + gradient Song et al., NeurIPS 2024 Score 0.766 Correct & Reconstruct →	0.766	0.865 38.33 dB / 0.983	0.742 30.56 dB / 0.925	0.691 27.94 dB / 0.880	✓ Certified	Song et al., NeurIPS 2024
5	DiffusionCT + gradient DiffusionCT + gradient Kazemi et al., ECCV 2024 Score 0.763 Correct & Reconstruct →	0.763	0.842 36.91 dB / 0.978	0.745 30.38 dB / 0.922	0.702 28.19 dB / 0.885	✓ Certified	Kazemi et al., ECCV 2024
6	DuDoTrans + gradient DuDoTrans + gradient Wang et al., MLMIR 2022 Score 0.742 Correct & Reconstruct →	0.742	0.818 35.07 dB / 0.968	0.723 28.83 dB / 0.897	0.686 27.64 dB / 0.873	✓ Certified	Wang et al., MLMIR 2022
7	Learned Primal-Dual + gradient Learned Primal-Dual + gradient Adler & Oktem, IEEE TMI 2018 Score 0.720 Correct & Reconstruct →	0.720	0.826 35.37 dB / 0.970	0.694 27.46 dB / 0.869	0.639 25.74 dB / 0.825	✓ Certified	Adler & Oktem, IEEE TMI 2018
8	RED-CNN + gradient RED-CNN + gradient Chen et al., IEEE TMI 2017 Score 0.714 Correct & Reconstruct →	0.714	0.786 32.21 dB / 0.945	0.689 27.0 dB / 0.858	0.666 26.43 dB / 0.844	✓ Certified	Chen et al., IEEE TMI 2017
9	PnP-ADMM + gradient PnP-ADMM + gradient Venkatakrishnan et al., IEEE GlobalSIP 2013 Score 0.704 Correct & Reconstruct →	0.704	0.748 30.33 dB / 0.922	0.690 27.98 dB / 0.880	0.673 26.29 dB / 0.840	✓ Certified	Venkatakrishnan et al., IEEE GlobalSIP 2013
10	PnP-DnCNN + gradient PnP-DnCNN + gradient Zhang et al., IEEE TIP 2017 Score 0.695 Correct & Reconstruct →	0.695	0.764 31.5 dB / 0.937	0.692 27.78 dB / 0.876	0.630 24.96 dB / 0.801	✓ Certified	Zhang et al., IEEE TIP 2017
11	FBPConvNet + gradient FBPConvNet + gradient Jin et al., IEEE TIP 2017 Score 0.672 Correct & Reconstruct →	0.672	0.793 33.32 dB / 0.955	0.663 25.62 dB / 0.821	0.560 21.94 dB / 0.687	✓ Certified	Jin et al., IEEE TIP 2017
12	TV-ADMM + gradient TV-ADMM + gradient Sidky et al., Phys. Med. Biol. 2008 Score 0.671 Correct & Reconstruct →	0.671	0.711 28.23 dB / 0.886	0.669 25.9 dB / 0.829	0.633 25.2 dB / 0.808	✓ Certified	Sidky et al., Phys. Med. Biol. 2008
13	FBP + gradient FBP + gradient Kak & Slaney, IEEE Press 1988 Score 0.612 Correct & Reconstruct →	0.612	0.643 24.57 dB / 0.788	0.621 24.34 dB / 0.780	0.573 22.64 dB / 0.717	✓ Certified	Kak & Slaney, IEEE Press 1988

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	Score-CT + gradient	0.865	38.33	0.983
2	DOLCE + gradient	0.846	36.69	0.977
3	DiffusionCT + gradient	0.842	36.91	0.978
4	CTFormer + gradient	0.840	36.9	0.978
5	CT-ViT + gradient	0.836	36.69	0.977
6	Learned Primal-Dual + gradient	0.826	35.37	0.97
7	DuDoTrans + gradient	0.818	35.07	0.968
8	FBPConvNet + gradient	0.793	33.32	0.955
9	RED-CNN + gradient	0.786	32.21	0.945
10	PnP-DnCNN + gradient	0.764	31.5	0.937
11	PnP-ADMM + gradient	0.748	30.33	0.922
12	TV-ADMM + gradient	0.711	28.23	0.886
13	FBP + gradient	0.643	24.57	0.788

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
source_dist	-5.0	10.0	mm
beam_hardening	-0.02	0.04
scatter	-0.05	0.1

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	CTFormer + gradient	0.800	35.44	0.97
2	CT-ViT + gradient	0.789	34.17	0.962
3	DOLCE + gradient	0.755	30.79	0.928
4	DiffusionCT + gradient	0.745	30.38	0.922
5	Score-CT + gradient	0.742	30.56	0.925
6	DuDoTrans + gradient	0.723	28.83	0.897
7	Learned Primal-Dual + gradient	0.694	27.46	0.869
8	PnP-DnCNN + gradient	0.692	27.78	0.876
9	PnP-ADMM + gradient	0.690	27.98	0.88
10	RED-CNN + gradient	0.689	27.0	0.858
11	TV-ADMM + gradient	0.669	25.9	0.829
12	FBPConvNet + gradient	0.663	25.62	0.821
13	FBP + gradient	0.621	24.34	0.78

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
source_dist	-6.0	9.0	mm
beam_hardening	-0.024	0.036
scatter	-0.06	0.09

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	CTFormer + gradient	0.770	33.68	0.958
2	CT-ViT + gradient	0.767	32.42	0.947
3	DOLCE + gradient	0.711	28.72	0.895
4	DiffusionCT + gradient	0.702	28.19	0.885
5	Score-CT + gradient	0.691	27.94	0.88
6	DuDoTrans + gradient	0.686	27.64	0.873
7	PnP-ADMM + gradient	0.673	26.29	0.84
8	RED-CNN + gradient	0.666	26.43	0.844
9	Learned Primal-Dual + gradient	0.639	25.74	0.825
10	TV-ADMM + gradient	0.633	25.2	0.808
11	PnP-DnCNN + gradient	0.630	24.96	0.801
12	FBP + gradient	0.573	22.64	0.717
13	FBPConvNet + gradient	0.560	21.94	0.687

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
source_dist	-3.5	11.5	mm
beam_hardening	-0.014	0.046
scatter	-0.035	0.115

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

Digital X-ray radiography produces a 2D projection image by transmitting X-rays through the body onto a flat-panel detector. The forward model follows Beer-Lambert attenuation: y = I_0 * exp(-integral(mu(s) ds)) + n where mu is the linear attenuation coefficient along each ray. The image is a superposition of all structures along the beam path. Primary degradations include quantum noise (Poisson), scatter, and geometric magnification artifacts.

Principle

X-ray radiography produces a 2-D projection image of the patient's internal structures by measuring the transmitted X-ray intensity after passing through the body. Dense structures (bone, metal) attenuate more X-rays and appear bright on the detector. The image represents the line-integral of the attenuation coefficient along each ray path.

How to Build the System

An X-ray tube (stationary or rotating anode, 40-150 kVp) produces a divergent beam. The patient stands or lies between the tube and a flat-panel detector (amorphous silicon with CsI scintillator, or amorphous selenium for direct conversion). Anti-scatter grid (Bucky grid) is placed before the detector. Automatic exposure control (AEC) sets mAs based on patient thickness. Calibration includes dark field, flatfield, and defective pixel mapping.

Common Reconstruction Algorithms

Flat-field correction (gain/offset normalization)
Logarithmic transform for linear attenuation mapping
Anti-scatter grid artifact removal
Dual-energy subtraction (bone/soft-tissue separation)
Deep-learning denoising for low-dose radiography

Common Mistakes

Under-exposure causing excessive quantum noise, especially in obese patients
Grid artifacts from misaligned anti-scatter grid
Patient motion blur in long-exposure radiographs
Incorrect windowing (display LUT) obscuring diagnostic information
Scatter radiation degrading image contrast in thick body parts

How to Avoid Mistakes

Use AEC and verify exposure indicator falls within acceptable range
Ensure grid is properly aligned with the X-ray focal spot distance
Use shortest possible exposure time; instruct patient to hold breath
Apply appropriate DICOM windowing presets for the anatomical region
Use an appropriate anti-scatter grid ratio (8:1 to 12:1) for thick body parts

Forward-Model Mismatch Cases

The widefield fallback applies additive Gaussian blur, but X-ray radiography follows Beer-Lambert attenuation: I = I_0 * exp(-integral(mu(x,y,z) dz)) — the exponential transmission model is fundamentally different from linear convolution
The Gaussian blur preserves mean intensity, but X-ray attenuation reduces intensity exponentially with material thickness and density — the fallback cannot model absorption contrast, bone/soft-tissue differentiation, or scatter

How to Correct the Mismatch

Use the X-ray radiography operator implementing Beer-Lambert transmission: y = I_0 * exp(-A*x) + scatter + noise, where A is the projection matrix along the beam direction
Include scatter rejection (anti-scatter grid model), detector response (DQE), and quantum noise (Poisson statistics) for physically accurate forward modeling

Experimental Setup — Signal Chain

Experimental setup diagram for X-ray Radiography

Reconstruction Gallery — 4 Scenes × 3 Scenarios

Method: CPU_baseline | Mismatch: nominal (nominal=True, perturbed=False)

I Ideal II Mismatch (uncorrected) III Oracle correction

I — Ideal

Ground Truth

Measurement

Reconstruction

14.80278336545788 dB SSIM 0.2898134486417238

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.687730656955978 dB SSIM 0.04575338765981969

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.768744450013651 dB SSIM 0.12684228475992015

I — Ideal

Ground Truth

Measurement

Reconstruction

15.045301424885293 dB SSIM 0.28567552083909603

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.697651814816277 dB SSIM 0.047738488895589425

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.91000303549352 dB SSIM 0.12823089632703247

I — Ideal

Ground Truth

Measurement

Reconstruction

14.990746783779773 dB SSIM 0.2846084456546437

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.683422700013802 dB SSIM 0.05057598958450131

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.95601267452069 dB SSIM 0.13121764093285918

I — Ideal

Ground Truth

Measurement

Reconstruction

14.689885976918243 dB SSIM 0.2938265882567135

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.670413526214027 dB SSIM 0.04754671710317928

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.905466417804053 dB SSIM 0.1298821443735687

Mean PSNR Across All Scenes

14.882179387760297 dB

I — Ideal

11.68480467450002 dB

II — Mismatch

14.885056644457979 dB

III — Oracle

Degradation: -3.2 dB Recovery: +3.2 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	14.80278336545788	0.2898134486417238	11.687730656955978	0.04575338765981969	14.768744450013651	0.12684228475992015
scene_01	15.045301424885293	0.28567552083909603	11.697651814816277	0.047738488895589425	14.91000303549352	0.12823089632703247
scene_02	14.990746783779773	0.2846084456546437	11.683422700013802	0.05057598958450131	14.95601267452069	0.13121764093285918
scene_03	14.689885976918243	0.2938265882567135	11.670413526214027	0.04754671710317928	14.905466417804053	0.1298821443735687
Mean	14.882179387760297	0.2884810008480443	11.68480467450002	0.04790364581077242	14.885056644457979	0.12904324159834513

Experimental Setup

Instrument: Carestream DRX-Evolution / Siemens Ysio Max

Image Size: 2048x2048

Pixel Pitch Mm: 0.1

Kvp: 120

Mas: 4.0

Sid Cm: 180

Detector: flat-panel (CsI + aSi TFT)

Dataset: CheXpert, MIMIC-CXR, NIH ChestX-ray14

Key References

Irvin et al., 'CheXpert: A large chest radiograph dataset', AAAI 2019
Wang et al., 'ChestX-ray8: Hospital-scale chest X-ray database', CVPR 2017

Canonical Datasets

CheXpert (Stanford, 224K studies)
MIMIC-CXR (MIT/BIDMC, 377K images)
NIH ChestX-ray14 (112K images)

Spec DAG — Forward Model Pipeline

Π(proj) → D(g, η₁)

Π X-ray Projection (proj)

D Flat-Panel Detector (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
Δd	source_dist	Source-to-detector distance error (mm)	5.0
β	beam_hardening	Beam-hardening coefficient	0.02
f_s	scatter	Scatter-to-primary ratio error	0.05

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.