CT

X-ray Computed Tomography

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
🥇	CT-FM CT-FM Wang 2026 44.1 dB SSIM 0.974 Checkpoint unavailable	0.972	44.1	0.974	✓ Certified	Wang 2026
🥈	PINER-CT PINER-CT Sun 2025 43.6 dB SSIM 0.970 Try in SpecLab →	0.962	43.6	0.970	✓ Certified	Sun 2025
🥉	CT-MAE CT-MAE Chen 2024 43.2 dB SSIM 0.968 Checkpoint unavailable	0.954	43.2	0.968	✓ Certified	Chen 2024
4	Score-CT Score-CT Gao 2024 42.8 dB SSIM 0.965 Checkpoint unavailable	0.946	42.8	0.965	✓ Certified	Gao 2024
5	DiffusionMBIR DiffusionMBIR Song 2024 42.5 dB SSIM 0.963 Checkpoint unavailable	0.940	42.5	0.963	✓ Certified	Song 2024
6	CTformer CTformer Wang 2023 41.2 dB SSIM 0.954 Checkpoint unavailable	0.914	41.2	0.954	✓ Certified	Wang 2023
7	Eformer Eformer Wang 2022 40.3 dB SSIM 0.948 Checkpoint unavailable	0.896	40.3	0.948	✓ Certified	Wang 2022
8	TransCT TransCT Xia 2021 39.8 dB SSIM 0.942 Checkpoint unavailable	0.884	39.8	0.942	✓ Certified	Xia 2021
9	DiffusionCT DiffusionCT 2305.18727 38.2 dB SSIM 0.965 Try in SpecLab →	0.869	38.2	0.965	↻ Reproduced	2305.18727
10	DuDoRNet DuDoRNet Zhou 2020 38.5 dB SSIM 0.931 Checkpoint unavailable	0.857	38.5	0.931	✓ Certified	Zhou 2020
11	iCT-Net iCT-Net Li 2019 37.5 dB SSIM 0.925 Checkpoint unavailable	0.838	37.5	0.925	✓ Certified	Li 2019
12	LEARN LEARN Chen 2018 36.8 dB SSIM 0.919 Checkpoint unavailable	0.823	36.8	0.919	✓ Certified	Chen 2018
13	RED-CNN RED-CNN Chen 2017 36.3 dB SSIM 0.914 Checkpoint unavailable	0.812	36.3	0.914	✓ Certified	Chen 2017
14	FBPConvNet FBPConvNet Jin 2017 34.1 dB SSIM 0.891 Checkpoint unavailable	0.764	34.1	0.891	✓ Certified	Jin 2017
15	WGAN-CT WGAN-CT Wolterink 2017 33.9 dB SSIM 0.887 Checkpoint unavailable	0.758	33.9	0.887	✓ Certified	Wolterink 2017
16	CT-U-Net CT-U-Net Han 2016 33.5 dB SSIM 0.883 Checkpoint unavailable	0.750	33.5	0.883	✓ Certified	Han 2016
17	PnP-ADMM PnP-ADMM Venkatakrishnan 2013 32.3 dB SSIM 0.868 Try in SpecLab →	0.722	32.3	0.868	✓ Certified	Venkatakrishnan 2013
18	DLCT DLCT Xu 2012 31.9 dB SSIM 0.862 Try in SpecLab →	0.713	31.9	0.862	✓ Certified	Xu 2012
19	BM3D-CT BM3D-CT Dabov 2007; Chen 2014 31.5 dB SSIM 0.856 Try in SpecLab →	0.703	31.5	0.856	✓ Certified	Dabov 2007; Chen 2014
20	TV-ADMM TV-ADMM Sidky 2008 30.4 dB SSIM 0.842 Try in SpecLab →	0.678	30.4	0.842	✓ Certified	Sidky 2008
21	ART-TV ART-TV Li 2004 29.8 dB SSIM 0.831 Try in SpecLab →	0.662	29.8	0.831	✓ Certified	Li 2004
22	SART SART Andersen 1984 28.7 dB SSIM 0.812 Try in SpecLab →	0.634	28.7	0.812	✓ Certified	Andersen 1984
23	OSEM OSEM Hudson 1994 27.5 dB SSIM 0.795 Try in SpecLab →	0.606	27.5	0.795	✓ Certified	Hudson 1994
24	CGLS CGLS Bjorck 1996 27.1 dB SSIM 0.788 Try in SpecLab →	0.596	27.1	0.788	✓ Certified	Bjorck 1996
25	FBP FBP Kak 1988 25.2 dB SSIM 0.771 Try in SpecLab →	0.555	25.2	0.771	✓ Certified	Kak 1988

Dataset: PWM Benchmark (24 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
🥇	CT-FM + gradient CT-FM + gradient Wang et al., Nature MI 2026 Score 0.844 Correct & Reconstruct →	0.844	0.910 42.34 dB / 0.992	0.828 35.87 dB / 0.973	0.795 33.58 dB / 0.958	✓ Certified	Wang et al., Nature MI 2026
🥈	Score-CT + gradient Score-CT + gradient Gao et al., IEEE TMI 2024 Score 0.824 Correct & Reconstruct →	0.824	0.877 40.38 dB / 0.989	0.803 35.65 dB / 0.972	0.791 33.62 dB / 0.958	✓ Certified	Gao et al., IEEE TMI 2024
🥉	CT-MAE + gradient CT-MAE + gradient Chen et al., MICCAI 2024 Score 0.823 Correct & Reconstruct →	0.823	0.902 41.95 dB / 0.992	0.803 35.01 dB / 0.968	0.765 32.26 dB / 0.945	✓ Certified	Chen et al., MICCAI 2024
4	TransCT + gradient TransCT + gradient Xia et al., MICCAI 2021 Score 0.815 Correct & Reconstruct →	0.815	0.865 38.72 dB / 0.984	0.798 34.88 dB / 0.967	0.782 33.47 dB / 0.957	✓ Certified	Xia et al., MICCAI 2021
5	CTformer + gradient CTformer + gradient Wang et al., MICCAI 2023 Score 0.815 Correct & Reconstruct →	0.815	0.861 39.08 dB / 0.985	0.814 36.23 dB / 0.975	0.771 33.11 dB / 0.954	✓ Certified	Wang et al., MICCAI 2023
6	PINER-CT + gradient PINER-CT + gradient Sun et al., CVPR 2025 Score 0.813 Correct & Reconstruct →	0.813	0.886 41.79 dB / 0.991	0.795 34.39 dB / 0.964	0.757 32.17 dB / 0.944	✓ Certified	Sun et al., CVPR 2025
7	DiffusionMBIR + gradient DiffusionMBIR + gradient Song et al., arXiv 2024 Score 0.794 Correct & Reconstruct →	0.794	0.894 41.38 dB / 0.991	0.756 32.07 dB / 0.943	0.731 29.52 dB / 0.909	✓ Certified	Song et al., arXiv 2024
8	Eformer + gradient Eformer + gradient Wang et al., AAAI 2022 Score 0.793 Correct & Reconstruct →	0.793	0.871 39.27 dB / 0.986	0.797 34.49 dB / 0.964	0.711 29.42 dB / 0.908	✓ Certified	Wang et al., AAAI 2022
9	DuDoRNet + gradient DuDoRNet + gradient Zhou et al., CVPR 2020 Score 0.778 Correct & Reconstruct →	0.778	0.849 37.14 dB / 0.979	0.767 31.41 dB / 0.936	0.718 29.92 dB / 0.916	✓ Certified	Zhou et al., CVPR 2020
10	iCT-Net + gradient iCT-Net + gradient Li et al., IEEE TMI 2019 Score 0.764 Correct & Reconstruct →	0.764	0.814 34.76 dB / 0.966	0.754 31.21 dB / 0.934	0.723 29.5 dB / 0.909	✓ Certified	Li et al., IEEE TMI 2019
11	LEARN + gradient LEARN + gradient Chen et al., IEEE TPAMI 2018 Score 0.761 Correct & Reconstruct →	0.761	0.831 35.77 dB / 0.972	0.740 30.6 dB / 0.926	0.711 28.35 dB / 0.888	✓ Certified	Chen et al., IEEE TPAMI 2018
12	PnP-ADMM + gradient PnP-ADMM + gradient Venkatakrishnan et al., GlobalSIP 2013 Score 0.725 Correct & Reconstruct →	0.725	0.747 30.38 dB / 0.922	0.722 29.47 dB / 0.908	0.706 28.55 dB / 0.892	✓ Certified	Venkatakrishnan et al., GlobalSIP 2013
13	BM3D-CT + gradient BM3D-CT + gradient Dabov et al., IEEE TIP 2007; Chen 2014 Score 0.704 Correct & Reconstruct →	0.704	0.754 29.97 dB / 0.916	0.697 28.04 dB / 0.882	0.660 25.85 dB / 0.828	✓ Certified	Dabov et al., IEEE TIP 2007; Chen 2014
14	RED-CNN + gradient RED-CNN + gradient Chen et al., IEEE TMI 2017 Score 0.684 Correct & Reconstruct →	0.684	0.801 33.88 dB / 0.960	0.654 25.48 dB / 0.817	0.597 23.18 dB / 0.738	✓ Certified	Chen et al., IEEE TMI 2017
15	DLCT + gradient DLCT + gradient Xu et al., IEEE TMI 2012 Score 0.675 Correct & Reconstruct →	0.675	0.762 30.6 dB / 0.926	0.666 26.34 dB / 0.841	0.598 23.95 dB / 0.767	✓ Certified	Xu et al., IEEE TMI 2012
16	WGAN-CT + gradient WGAN-CT + gradient Wolterink et al., IEEE TMI 2017 Score 0.665 Correct & Reconstruct →	0.665	0.767 31.61 dB / 0.938	0.635 24.58 dB / 0.789	0.593 23.42 dB / 0.747	✓ Certified	Wolterink et al., IEEE TMI 2017
17	SART + gradient SART + gradient Andersen & Kak, Ultrason. Imaging 1984 Score 0.664 Correct & Reconstruct →	0.664	0.679 26.38 dB / 0.842	0.660 25.58 dB / 0.820	0.652 25.88 dB / 0.829	✓ Certified	Andersen & Kak, Ultrason. Imaging 1984
18	CT-U-Net + gradient CT-U-Net + gradient Han et al., Phys. Med. Biol. 2016 Score 0.653 Correct & Reconstruct →	0.653	0.763 31.3 dB / 0.935	0.621 24.74 dB / 0.794	0.575 23.02 dB / 0.732	✓ Certified	Han et al., Phys. Med. Biol. 2016
19	FBPConvNet + gradient FBPConvNet + gradient Jin et al., IEEE TMI 2017 Score 0.647 Correct & Reconstruct →	0.647	0.774 32.33 dB / 0.946	0.605 23.82 dB / 0.762	0.563 22.1 dB / 0.694	✓ Certified	Jin et al., IEEE TMI 2017
20	CGLS + gradient CGLS + gradient Bjorck, SIAM 1996 Score 0.643 Correct & Reconstruct →	0.643	0.650 25.33 dB / 0.812	0.636 25.13 dB / 0.806	0.643 25.35 dB / 0.813	✓ Certified	Bjorck, SIAM 1996
21	OSEM + gradient OSEM + gradient Hudson & Larkin, IEEE TMI 1994 Score 0.619 Correct & Reconstruct →	0.619	0.654 25.31 dB / 0.812	0.631 24.83 dB / 0.797	0.573 23.02 dB / 0.732	✓ Certified	Hudson & Larkin, IEEE TMI 1994
22	TV-ADMM + gradient TV-ADMM + gradient Sidky & Pan, Phys. Med. Biol. 2008 Score 0.587 Correct & Reconstruct →	0.587	0.736 28.86 dB / 0.898	0.552 21.34 dB / 0.661	0.473 19.37 dB / 0.568	✓ Certified	Sidky & Pan, Phys. Med. Biol. 2008
23	ART-TV + gradient ART-TV + gradient Li et al., Med. Phys. 2004 Score 0.556 Correct & Reconstruct →	0.556	0.697 27.28 dB / 0.865	0.517 20.59 dB / 0.627	0.453 18.02 dB / 0.501	✓ Certified	Li et al., Med. Phys. 2004
24	FBP + gradient FBP + gradient Kak & Slaney, IEEE Press 1988 Score 0.537 Correct & Reconstruct →	0.537	0.588 22.38 dB / 0.706	0.533 20.74 dB / 0.634	0.489 19.92 dB / 0.595	✓ 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 11 scenes

Full-access tier: 11 real patient CT slices from LoDoPaB-CT (LIDC/IDRI test split).

What you get & how to use

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

How to use: Load ct_challenge_public.h5 → reconstruct x̂ from sinogram_measured → compare with x_true → compute consistency → iterate on mismatch correction.

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

Public Leaderboard

#	Method	Score	PSNR	SSIM
1	CT-FM + gradient	0.910	42.34	0.992
2	CT-MAE + gradient	0.902	41.95	0.992
3	DiffusionMBIR + gradient	0.894	41.38	0.991
4	PINER-CT + gradient	0.886	41.79	0.991
5	Score-CT + gradient	0.877	40.38	0.989
6	Eformer + gradient	0.871	39.27	0.986
7	TransCT + gradient	0.865	38.72	0.984
8	CTformer + gradient	0.861	39.08	0.985
9	DuDoRNet + gradient	0.849	37.14	0.979
10	LEARN + gradient	0.831	35.77	0.972
11	iCT-Net + gradient	0.814	34.76	0.966
12	RED-CNN + gradient	0.801	33.88	0.96
13	FBPConvNet + gradient	0.774	32.33	0.946
14	WGAN-CT + gradient	0.767	31.61	0.938
15	CT-U-Net + gradient	0.763	31.3	0.935
16	DLCT + gradient	0.762	30.6	0.926
17	BM3D-CT + gradient	0.754	29.97	0.916
18	PnP-ADMM + gradient	0.747	30.38	0.922
19	TV-ADMM + gradient	0.736	28.86	0.898
20	ART-TV + gradient	0.697	27.28	0.865
21	SART + gradient	0.679	26.38	0.842
22	OSEM + gradient	0.654	25.31	0.812
23	CGLS + gradient	0.650	25.33	0.812
24	FBP + gradient	0.588	22.38	0.706

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
center_offset_px	-4.0	6.0	px
angle_error_deg	-6.5	9.5	deg
beam_hardening_beta	-0.1	0.2
detector_tilt_deg	-2.5	3.5	deg

View Details →

Dev 20 scenes

Blind evaluation: 20 real patient CT slices from LoDoPaB-CT (validation split, patients 0–63).

What you get & how to use

What you get: Measured sinogram (y), ideal forward operator (H), and spec ranges. No ground truth.

How to use: Apply your pipeline from Public tier. Self-check via consistency metric. Ground truth scored server-side.

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

Dev Leaderboard

#	Method	Score	PSNR	SSIM
1	CT-FM + gradient	0.828	35.87	0.973
2	CTformer + gradient	0.814	36.23	0.975
3	Score-CT + gradient	0.803	35.65	0.972
4	CT-MAE + gradient	0.803	35.01	0.968
5	TransCT + gradient	0.798	34.88	0.967
6	Eformer + gradient	0.797	34.49	0.964
7	PINER-CT + gradient	0.795	34.39	0.964
8	DuDoRNet + gradient	0.767	31.41	0.936
9	DiffusionMBIR + gradient	0.756	32.07	0.943
10	iCT-Net + gradient	0.754	31.21	0.934
11	LEARN + gradient	0.740	30.6	0.926
12	PnP-ADMM + gradient	0.722	29.47	0.908
13	BM3D-CT + gradient	0.697	28.04	0.882
14	DLCT + gradient	0.666	26.34	0.841
15	SART + gradient	0.660	25.58	0.82
16	RED-CNN + gradient	0.654	25.48	0.817
17	CGLS + gradient	0.636	25.13	0.806
18	WGAN-CT + gradient	0.635	24.58	0.789
19	OSEM + gradient	0.631	24.83	0.797
20	CT-U-Net + gradient	0.621	24.74	0.794
21	FBPConvNet + gradient	0.605	23.82	0.762
22	TV-ADMM + gradient	0.552	21.34	0.661
23	FBP + gradient	0.533	20.74	0.634
24	ART-TV + gradient	0.517	20.59	0.627

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
center_offset_px	-3.0	7.0	px
angle_error_deg	-5.0	11.0	deg
beam_hardening_beta	-0.07	0.23
detector_tilt_deg	-2.0	4.0	deg

Submit Reconstruction View Details →

Hidden 20 scenes

Fully blind: 20 real LoDoPaB-CT slices (validation split, patients 64–127) with adversarial modifications (metal inserts, lesions, calcifications).

What you get & how to use

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

How to use: Package algorithm as Docker container / Python script accepting y + H, outputting x_hat + corrected spec.

What to submit: Containerized algorithm. Scored server-side against adversarial phantoms.

Hidden Leaderboard

#	Method	Score	PSNR	SSIM
1	CT-FM + gradient	0.795	33.58	0.958
2	Score-CT + gradient	0.791	33.62	0.958
3	TransCT + gradient	0.782	33.47	0.957
4	CTformer + gradient	0.771	33.11	0.954
5	CT-MAE + gradient	0.765	32.26	0.945
6	PINER-CT + gradient	0.757	32.17	0.944
7	DiffusionMBIR + gradient	0.731	29.52	0.909
8	iCT-Net + gradient	0.723	29.5	0.909
9	DuDoRNet + gradient	0.718	29.92	0.916
10	Eformer + gradient	0.711	29.42	0.908
11	LEARN + gradient	0.711	28.35	0.888
12	PnP-ADMM + gradient	0.706	28.55	0.892
13	BM3D-CT + gradient	0.660	25.85	0.828
14	SART + gradient	0.652	25.88	0.829
15	CGLS + gradient	0.643	25.35	0.813
16	DLCT + gradient	0.598	23.95	0.767
17	RED-CNN + gradient	0.597	23.18	0.738
18	WGAN-CT + gradient	0.593	23.42	0.747
19	CT-U-Net + gradient	0.575	23.02	0.732
20	OSEM + gradient	0.573	23.02	0.732
21	FBPConvNet + gradient	0.563	22.1	0.694
22	FBP + gradient	0.489	19.92	0.595
23	TV-ADMM + gradient	0.473	19.37	0.568
24	ART-TV + gradient	0.453	18.02	0.501

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
center_offset_px	-1.0	9.0	px
angle_error_deg	-2.0	14.0	deg
beam_hardening_beta	0.07	0.37
detector_tilt_deg	-0.5	5.5	deg

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

X-ray CT reconstructs cross-sectional images from a set of line-integral projections (sinogram) acquired as an X-ray source and detector array rotate around the patient. The forward model is the Radon transform: y = R*x + n where R computes line integrals along each ray. Sparse-view and low-dose protocols reduce radiation but introduce streak artifacts and noise. Reconstruction uses filtered back-projection (FBP) or iterative methods (MBIR, DL post-processing).

Principle

X-ray Computed Tomography reconstructs cross-sectional images from multiple X-ray projection measurements acquired at different angles around the patient. The Beer-Lambert law governs X-ray attenuation: I = I₀ exp(-∫μ(x,y) dl), and the Radon transform relates projections to the attenuation map. Filtered back-projection or iterative algorithms invert the Radon transform to produce volumetric images.

How to Build the System

A clinical CT scanner consists of a rotating gantry with an X-ray tube (80-140 kVp, 50-800 mA) and a curved detector array (64-320 rows of scintillator-photodiode elements) on opposing sides. The gantry rotates at 0.25-0.5 s per revolution. Helical scanning moves the patient table continuously through the gantry. Key calibrations: air scans, detector gain normalization, beam-hardening correction LUTs, and geometric calibration.

Common Reconstruction Algorithms

Filtered back-projection (FBP) with Ram-Lak or Shepp-Logan filter
FDK (Feldkamp-Davis-Kress) for cone-beam geometry
Iterative reconstruction: SART, OS-SIRT
Model-based iterative reconstruction (MBIR) with statistical noise model
Deep-learning reconstruction (FBPConvNet, LEARN, WGAN-VGG for low-dose CT)

Common Mistakes

Ring artifacts from uncorrected detector gain variations
Beam-hardening artifacts (cupping, streaks near bone/metal) not corrected
Patient motion during scan causing blurring and streaks
Insufficient angular sampling producing streak or aliasing artifacts
Metal artifacts from implants overwhelming reconstruction algorithms

How to Avoid Mistakes

Perform regular air calibrations and detector flatfield corrections
Apply polynomial beam-hardening correction or dual-energy decomposition
Use gating (cardiac/respiratory) or fast rotation to reduce motion artifacts
Ensure adequate number of projections (≥ π × detector columns for FBP)
Use metal artifact reduction algorithms (MAR, iterative forward-projection inpainting)

Forward-Model Mismatch Cases

The widefield fallback produces a blurred (64,64) image, but CT acquires a sinogram of shape (180,64) via the Radon transform (line integrals at multiple angles) — any reconstruction algorithm expecting sinogram input will crash
The Gaussian blur preserves spatial structure, but the Radon transform converts spatial information into angular projections — the fallback output bears no physical relationship to X-ray transmission measurements

How to Correct the Mismatch

Use the CT operator implementing the discrete Radon transform: y(theta,s) = integral of f(x,y) along line at angle theta and offset s, producing a (n_angles, n_detectors) sinogram
Reconstruct using filtered back-projection (FBP) or iterative algorithms (SART, ADMM-TV) that require the correct Radon transform / back-projection pair

Experimental Setup — Signal Chain

Experimental setup diagram for X-ray Computed Tomography

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

13.015083310309393 dB SSIM 0.11002121493463184

II — Mismatch

Ground Truth

Measurement

Reconstruction

10.897295334470806 dB SSIM 0.0538311684488209

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.55309786198268 dB SSIM 0.16300192959603202

I — Ideal

Ground Truth

Measurement

Reconstruction

13.626949465678774 dB SSIM 0.10592147588153143

II — Mismatch

Ground Truth

Measurement

Reconstruction

10.913853966202094 dB SSIM 0.0519764645569369

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.525584396667453 dB SSIM 0.1500916901726949

I — Ideal

Ground Truth

Measurement

Reconstruction

14.796126301820475 dB SSIM 0.11689158958981596

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.751592440750972 dB SSIM 0.05612148758089694

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

15.249338245235649 dB SSIM 0.15605377346516436

I — Ideal

Ground Truth

Measurement

Reconstruction

14.343204740716882 dB SSIM 0.10873706336879811

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.35862568759396 dB SSIM 0.054887129803128966

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

14.958310096170258 dB SSIM 0.15856494120853848

Mean PSNR Across All Scenes

13.94534095463138 dB

I — Ideal

11.230341857254459 dB

II — Mismatch

14.82158265001401 dB

III — Oracle

Degradation: -2.7 dB Recovery: +3.6 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	13.015083310309393	0.11002121493463184	10.897295334470806	0.0538311684488209	14.55309786198268	0.16300192959603202
scene_01	13.626949465678774	0.10592147588153143	10.913853966202094	0.0519764645569369	14.525584396667453	0.1500916901726949
scene_02	14.796126301820475	0.11689158958981596	11.751592440750972	0.05612148758089694	15.249338245235649	0.15605377346516436
scene_03	14.343204740716882	0.10873706336879811	11.35862568759396	0.054887129803128966	14.958310096170258	0.15856494120853848
Mean	13.94534095463138	0.11039283594369434	11.230341857254459	0.05420406259744592	14.82158265001401	0.15692808361060745

Experimental Setup

Instrument: Siemens SOMATOM Force / GE Revolution CT

Image Size: 512x512

Num Views: 60

Full Dose Views: 1000

Detector Pixels: 736

Kvp: 120

Dose Level: 25% of full dose (quarter-dose)

Reconstruction: FBP + DL denoising / end-to-end

Dataset: LoDoPaB-CT, DeepLesion

Key References

Feldkamp et al., 'Practical cone-beam algorithm', J. Opt. Soc. Am. A 1, 612-619 (1984)
Leuschner et al., 'LoDoPaB-CT, a benchmark dataset for low-dose CT reconstruction', Scientific Data 8, 109 (2021)

Canonical Datasets

LoDoPaB-CT (Scientific Data 2021)
DeepLesion (NIH Clinical Center)
AAPM Low-Dose CT Grand Challenge

Spec DAG — Forward Model Pipeline

R(θ) → Π(fan) → D(g, η₁)

R Gantry Rotation (θ)

Π Fan-Beam Projection (fan)

D Detector (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
Δc	center_offset	Center-of-rotation offset (pixels)	1.5
Δθ	angle_error	Gantry angle error (deg)	0.5
β	beam_hardening	Beam-hardening coefficient	0.03
φ	detector_tilt	Detector tilt (deg)	0.2

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.