Confocal Laser Endomicroscopy (CLE)

Confocal Laser Endomicroscopy (CLE)

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
🥇 DiffusionEndo 0.887 39.4 0.960 ✓ Certified Li 2024
🥈 Restormer-CLE 0.859 38.1 0.949 ✓ Certified Zamir 2022
🥉 SwinIR-CLE 0.831 36.8 0.936 ✓ Certified Liang 2021
4 PINN-CLE 0.817 36.1 0.930 ✓ Certified Kang 2022
5 CARE-CLE 0.797 35.2 0.920 ✓ Certified Weigert 2018
6 U-Net-CLE 0.764 33.8 0.902 ✓ Certified Andre 2011
7 DnCNN-CLE 0.704 31.2 0.868 ✓ Certified Zhang 2017
8 BM3D-CLE 0.621 27.8 0.815 ✓ Certified Dabov 2007
9 NLM-Speckle 0.562 25.5 0.775 ✓ Certified Buades 2005

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
🥇 DiffusionEndo + gradient 0.785
0.840
37.39 dB / 0.980
0.769
33.03 dB / 0.953
0.746
31.65 dB / 0.939
✓ Certified Li et al., Med. Image Anal. 2024
🥈 Restormer-CLE + gradient 0.784
0.825
35.96 dB / 0.973
0.793
34.34 dB / 0.963
0.734
31.22 dB / 0.934
✓ Certified Zamir et al., CVPR 2022 (CLE)
🥉 SwinIR-CLE + gradient 0.728
0.829
35.62 dB / 0.971
0.714
29.23 dB / 0.904
0.642
25.85 dB / 0.828
✓ Certified Liang et al., ICCV 2021 (CLE)
4 CARE-CLE + gradient 0.713
0.783
32.27 dB / 0.946
0.708
28.56 dB / 0.892
0.648
25.23 dB / 0.809
✓ Certified Weigert et al., Nat. Methods 2018 (CLE)
5 PINN-CLE + gradient 0.687
0.821
34.81 dB / 0.966
0.656
26.16 dB / 0.836
0.585
22.73 dB / 0.720
✓ Certified Kang et al., Med. Phys. 2022
6 U-Net-CLE + gradient 0.653
0.791
32.58 dB / 0.949
0.614
24.45 dB / 0.784
0.553
22.2 dB / 0.698
✓ Certified Andre et al., Med. Image Anal. 2011 (updated DL)
7 BM3D-CLE + gradient 0.651
0.665
26.02 dB / 0.833
0.671
26.48 dB / 0.845
0.617
24.85 dB / 0.797
✓ Certified Dabov et al., IEEE TIP 2007
8 DnCNN-CLE + gradient 0.651
0.723
28.57 dB / 0.892
0.649
25.02 dB / 0.803
0.581
22.73 dB / 0.720
✓ Certified Zhang et al., IEEE TIP 2017
9 NLM-Speckle + gradient 0.597
0.609
23.49 dB / 0.750
0.610
23.76 dB / 0.760
0.573
22.62 dB / 0.716
✓ Certified Buades et al., CVPR 2005

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 DiffusionEndo + gradient 0.840 37.39 0.98
2 SwinIR-CLE + gradient 0.829 35.62 0.971
3 Restormer-CLE + gradient 0.825 35.96 0.973
4 PINN-CLE + gradient 0.821 34.81 0.966
5 U-Net-CLE + gradient 0.791 32.58 0.949
6 CARE-CLE + gradient 0.783 32.27 0.946
7 DnCNN-CLE + gradient 0.723 28.57 0.892
8 BM3D-CLE + gradient 0.665 26.02 0.833
9 NLM-Speckle + gradient 0.609 23.49 0.75
Spec Ranges (3 parameters)
Parameter Min Max Unit
fiber_bundle_honeycomb_pattern -0.15 0.15 -
motion_artifact -2.0 4.0 px/frame
fluorescein_concentration_variation 0.6 1.8 relative
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 Restormer-CLE + gradient 0.793 34.34 0.963
2 DiffusionEndo + gradient 0.769 33.03 0.953
3 SwinIR-CLE + gradient 0.714 29.23 0.904
4 CARE-CLE + gradient 0.708 28.56 0.892
5 BM3D-CLE + gradient 0.671 26.48 0.845
6 PINN-CLE + gradient 0.656 26.16 0.836
7 DnCNN-CLE + gradient 0.649 25.02 0.803
8 U-Net-CLE + gradient 0.614 24.45 0.784
9 NLM-Speckle + gradient 0.610 23.76 0.76
Spec Ranges (3 parameters)
Parameter Min Max Unit
fiber_bundle_honeycomb_pattern -0.15 0.15 -
motion_artifact -2.4 3.6 px/frame
fluorescein_concentration_variation 0.52 1.72 relative
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 DiffusionEndo + gradient 0.746 31.65 0.939
2 Restormer-CLE + gradient 0.734 31.22 0.934
3 CARE-CLE + gradient 0.648 25.23 0.809
4 SwinIR-CLE + gradient 0.642 25.85 0.828
5 BM3D-CLE + gradient 0.617 24.85 0.797
6 PINN-CLE + gradient 0.585 22.73 0.72
7 DnCNN-CLE + gradient 0.581 22.73 0.72
8 NLM-Speckle + gradient 0.573 22.62 0.716
9 U-Net-CLE + gradient 0.553 22.2 0.698
Spec Ranges (3 parameters)
Parameter Min Max Unit
fiber_bundle_honeycomb_pattern -0.15 0.15 -
motion_artifact -1.4 4.6 px/frame
fluorescein_concentration_variation 0.72 1.92 relative
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 → C → D

M Modulation
C Convolution
D Detector

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
f_b fiber_bundle_honeycomb_pattern Fiber bundle honeycomb pattern (-) 0.0 0.0
m_a motion_artifact Motion artifact (px/frame) 0.0 2.0
f_c fluorescein_concentration_variation Fluorescein concentration variation (relative) 1.0 1.4

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.