CEST MRI

CEST MRI

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
🥇 DiffusionCEST 0.892 39.7 0.961 ✓ Certified Chen 2024
🥈 PromptCEST 0.869 38.6 0.951 ✓ Certified Liu 2024
🥉 CESTFormer 0.843 37.4 0.940 ✓ Certified Wu 2023
4 PINN-CEST 0.811 35.9 0.925 ✓ Certified Cohen 2022
5 U-Net-CEST 0.786 34.8 0.912 ✓ Certified Zhao 2021
6 DnCNN-CEST 0.724 32.1 0.878 ✓ Certified Zhang 2017
7 WASSR 0.640 28.5 0.831 ✓ Certified Kim 2009
8 Lorentzian-Fit 0.607 27.2 0.808 ✓ Certified Zaiss 2013
9 MTR-asym 0.544 24.8 0.761 ✓ Certified Zhou 2003

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
🥇 PromptCEST + gradient 0.771
0.830
36.35 dB / 0.975
0.770
32.52 dB / 0.948
0.713
28.35 dB / 0.888
✓ Certified Liu et al., MRM 2024
🥈 CESTFormer + gradient 0.758
0.817
35.62 dB / 0.971
0.755
30.94 dB / 0.930
0.702
27.86 dB / 0.878
✓ Certified Wu et al., IEEE TMI 2023
🥉 DiffusionCEST + gradient 0.747
0.843
37.67 dB / 0.981
0.715
29.51 dB / 0.909
0.683
27.72 dB / 0.875
✓ Certified Chen et al., NeurIPS 2024
4 PINN-CEST + gradient 0.739
0.793
33.19 dB / 0.954
0.724
28.89 dB / 0.898
0.700
27.64 dB / 0.873
✓ Certified Cohen et al., MRM 2022
5 U-Net-CEST + gradient 0.695
0.783
32.82 dB / 0.951
0.681
27.22 dB / 0.863
0.620
24.54 dB / 0.787
✓ Certified Zhao et al., MRM 2021
6 WASSR + gradient 0.662
0.668
25.68 dB / 0.823
0.661
26.3 dB / 0.840
0.657
26.12 dB / 0.835
✓ Certified Kim et al., MRM 2009
7 Lorentzian-Fit + gradient 0.640
0.652
25.43 dB / 0.815
0.641
25.29 dB / 0.811
0.627
24.75 dB / 0.794
✓ Certified Zaiss & Bachert, NMR Biomed. 2013
8 DnCNN-CEST + gradient 0.611
0.739
29.62 dB / 0.911
0.596
23.15 dB / 0.737
0.497
19.77 dB / 0.588
✓ Certified Zhang et al., IEEE TIP 2017 (CEST adapted)
9 MTR-asym + gradient 0.570
0.590
22.67 dB / 0.718
0.578
22.08 dB / 0.693
0.543
21.04 dB / 0.647
✓ Certified Zhou et al., Nat. Med. 2003

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 DiffusionCEST + gradient 0.843 37.67 0.981
2 PromptCEST + gradient 0.830 36.35 0.975
3 CESTFormer + gradient 0.817 35.62 0.971
4 PINN-CEST + gradient 0.793 33.19 0.954
5 U-Net-CEST + gradient 0.783 32.82 0.951
6 DnCNN-CEST + gradient 0.739 29.62 0.911
7 WASSR + gradient 0.668 25.68 0.823
8 Lorentzian-Fit + gradient 0.652 25.43 0.815
9 MTR-asym + gradient 0.590 22.67 0.718
Spec Ranges (4 parameters)
Parameter Min Max Unit
b0_inhomogeneity -10.0 20.0 Hz
b1_inhomogeneity -4.0 8.0 -
saturation_power_error -2.0 4.0 -
mt_contamination -6.0 12.0 -
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 PromptCEST + gradient 0.770 32.52 0.948
2 CESTFormer + gradient 0.755 30.94 0.93
3 PINN-CEST + gradient 0.724 28.89 0.898
4 DiffusionCEST + gradient 0.715 29.51 0.909
5 U-Net-CEST + gradient 0.681 27.22 0.863
6 WASSR + gradient 0.661 26.3 0.84
7 Lorentzian-Fit + gradient 0.641 25.29 0.811
8 DnCNN-CEST + gradient 0.596 23.15 0.737
9 MTR-asym + gradient 0.578 22.08 0.693
Spec Ranges (4 parameters)
Parameter Min Max Unit
b0_inhomogeneity -12.0 18.0 Hz
b1_inhomogeneity -4.8 7.2 -
saturation_power_error -2.4 3.6 -
mt_contamination -7.2 10.8 -
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 PromptCEST + gradient 0.713 28.35 0.888
2 CESTFormer + gradient 0.702 27.86 0.878
3 PINN-CEST + gradient 0.700 27.64 0.873
4 DiffusionCEST + gradient 0.683 27.72 0.875
5 WASSR + gradient 0.657 26.12 0.835
6 Lorentzian-Fit + gradient 0.627 24.75 0.794
7 U-Net-CEST + gradient 0.620 24.54 0.787
8 MTR-asym + gradient 0.543 21.04 0.647
9 DnCNN-CEST + gradient 0.497 19.77 0.588
Spec Ranges (4 parameters)
Parameter Min Max Unit
b0_inhomogeneity -7.0 23.0 Hz
b1_inhomogeneity -2.8 9.2 -
saturation_power_error -1.4 4.6 -
mt_contamination -4.2 13.8 -
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

Spec DAG — Forward Model Pipeline

M → F → S → D

M Modulation
F Fourier
S Sampling
D Detector

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
b_i b0_inhomogeneity B0 inhomogeneity (Hz) 0.0 10.0
b_i b1_inhomogeneity B1 inhomogeneity (-) 0.0 4.0
s_p saturation_power_error Saturation power error (-) 0.0 2.0
m_c mt_contamination MT contamination (-) 0.0 6.0

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.