Confocal Live-Cell

Confocal Live-Cell Microscopy

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
🥇	DiffusionCell DiffusionCell Gao 2024 39.2 dB SSIM 0.959 Checkpoint unavailable	0.883	39.2	0.959	✓ Certified	Gao 2024
🥈	Restormer-Micro Restormer-Micro Zamir 2022 37.8 dB SSIM 0.946 Checkpoint unavailable	0.853	37.8	0.946	✓ Certified	Zamir 2022
🥉	SwinIR-LiveCell SwinIR-LiveCell Liang 2021 36.2 dB SSIM 0.931 Checkpoint unavailable	0.819	36.2	0.931	✓ Certified	Liang 2021
4	CARE CARE Weigert 2018 33.5 dB SSIM 0.891 Checkpoint unavailable	0.754	33.5	0.891	✓ Certified	Weigert 2018
5	PN2V PN2V Krull 2020 32.9 dB SSIM 0.882 Checkpoint unavailable	0.739	32.9	0.882	✓ Certified	Krull 2020
6	Noise2Void Noise2Void Krull 2019 31.8 dB SSIM 0.871 Checkpoint unavailable	0.716	31.8	0.871	✓ Certified	Krull 2019
7	Noise2Self Noise2Self Batson 2019 30.5 dB SSIM 0.858 Checkpoint unavailable	0.687	30.5	0.858	✓ Certified	Batson 2019
8	NLM-Fluorescence NLM-Fluorescence Buades 2005 26.8 dB SSIM 0.795 Try in SpecLab →	0.594	26.8	0.795	✓ Certified	Buades 2005
9	VST-Denoise VST-Denoise Anscombe 1948 24.2 dB SSIM 0.751 Try in SpecLab →	0.529	24.2	0.751	✓ Certified	Anscombe 1948

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
🥇	DiffusionCell + gradient DiffusionCell + gradient Gao et al., Nat. Methods 2024 Score 0.765 Correct & Reconstruct →	0.765	0.857 37.54 dB / 0.980	0.745 31.2 dB / 0.933	0.694 28.42 dB / 0.889	✓ Certified	Gao et al., Nat. Methods 2024
🥈	Restormer-Micro + gradient Restormer-Micro + gradient Zamir et al., CVPR 2022 (microscopy) Score 0.758 Correct & Reconstruct →	0.758	0.841 36.47 dB / 0.976	0.755 31.19 dB / 0.933	0.678 26.52 dB / 0.846	✓ Certified	Zamir et al., CVPR 2022 (microscopy)
🥉	SwinIR-LiveCell + gradient SwinIR-LiveCell + gradient Liang et al., ICCV 2021 (live-cell) Score 0.742 Correct & Reconstruct →	0.742	0.824 35.19 dB / 0.969	0.733 30.13 dB / 0.919	0.669 26.67 dB / 0.850	✓ Certified	Liang et al., ICCV 2021 (live-cell)
4	PN2V + gradient PN2V + gradient Krull et al., ECCV 2020 Score 0.655 Correct & Reconstruct →	0.655	0.775 31.29 dB / 0.935	0.616 23.69 dB / 0.757	0.574 22.67 dB / 0.718	✓ Certified	Krull et al., ECCV 2020
5	Noise2Void + gradient Noise2Void + gradient Krull et al., CVPR 2019 Score 0.639 Correct & Reconstruct →	0.639	0.738 29.81 dB / 0.914	0.605 24.05 dB / 0.770	0.575 22.26 dB / 0.701	✓ Certified	Krull et al., CVPR 2019
6	CARE + gradient CARE + gradient Weigert et al., Nat. Methods 2018 Score 0.633 Correct & Reconstruct →	0.633	0.759 30.62 dB / 0.926	0.592 22.89 dB / 0.727	0.547 21.55 dB / 0.670	✓ Certified	Weigert et al., Nat. Methods 2018
7	NLM-Fluorescence + gradient NLM-Fluorescence + gradient Buades et al., CVPR 2005 Score 0.583 Correct & Reconstruct →	0.583	0.640 24.8 dB / 0.796	0.564 21.85 dB / 0.684	0.545 21.02 dB / 0.647	✓ Certified	Buades et al., CVPR 2005
8	Noise2Self + gradient Noise2Self + gradient Batson & Royer, ICML 2019 Score 0.573 Correct & Reconstruct →	0.573	0.707 27.61 dB / 0.872	0.527 21.09 dB / 0.650	0.486 19.4 dB / 0.570	✓ Certified	Batson & Royer, ICML 2019
9	VST-Denoise + gradient VST-Denoise + gradient Anscombe, Biometrika 1948 Score 0.534 Correct & Reconstruct →	0.534	0.603 22.71 dB / 0.720	0.520 20.16 dB / 0.606	0.478 19.34 dB / 0.567	✓ Certified	Anscombe, Biometrika 1948

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 5 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	DiffusionCell + gradient	0.857	37.54	0.98
2	Restormer-Micro + gradient	0.841	36.47	0.976
3	SwinIR-LiveCell + gradient	0.824	35.19	0.969
4	PN2V + gradient	0.775	31.29	0.935
5	CARE + gradient	0.759	30.62	0.926
6	Noise2Void + gradient	0.738	29.81	0.914
7	Noise2Self + gradient	0.707	27.61	0.872
8	NLM-Fluorescence + gradient	0.640	24.8	0.796
9	VST-Denoise + gradient	0.603	22.71	0.72

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
pinhole	-5.0	10.0	μm
refractive_index	1.51	1.525
photobleaching	-5.0	10.0	%

View Details →

Dev 5 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-Micro + gradient	0.755	31.19	0.933
2	DiffusionCell + gradient	0.745	31.2	0.933
3	SwinIR-LiveCell + gradient	0.733	30.13	0.919
4	PN2V + gradient	0.616	23.69	0.757
5	Noise2Void + gradient	0.605	24.05	0.77
6	CARE + gradient	0.592	22.89	0.727
7	NLM-Fluorescence + gradient	0.564	21.85	0.684
8	Noise2Self + gradient	0.527	21.09	0.65
9	VST-Denoise + gradient	0.520	20.16	0.606

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
pinhole	-6.0	9.0	μm
refractive_index	1.509	1.524
photobleaching	-6.0	9.0	%

Submit Reconstruction View Details →

Hidden 5 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	DiffusionCell + gradient	0.694	28.42	0.889
2	Restormer-Micro + gradient	0.678	26.52	0.846
3	SwinIR-LiveCell + gradient	0.669	26.67	0.85
4	Noise2Void + gradient	0.575	22.26	0.701
5	PN2V + gradient	0.574	22.67	0.718
6	CARE + gradient	0.547	21.55	0.67
7	NLM-Fluorescence + gradient	0.545	21.02	0.647
8	Noise2Self + gradient	0.486	19.4	0.57
9	VST-Denoise + gradient	0.478	19.34	0.567

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
pinhole	-3.5	11.5	μm
refractive_index	1.5115	1.5265
photobleaching	-3.5	11.5	%

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

Laser scanning confocal microscopy for live-cell imaging. A focused laser scans the specimen point by point, and a pinhole rejects out-of-focus light. The image formation is modelled as convolution with the confocal PSF (product of excitation and detection PSFs). Fast acquisition rates for live cells often sacrifice SNR due to short pixel dwell times. Reconstruction involves deconvolution with the confocal PSF and temporal denoising across frames.

Principle

A focused laser spot is scanned across the specimen and a pinhole in front of the detector rejects out-of-focus fluorescence, providing optical sectioning. The image formation is modeled as a point-by-point convolution with the confocal PSF (product of excitation and detection PSFs). For live-cell work, speed and gentleness are prioritized.

How to Build the System

Equip a laser-scanning confocal head (e.g., Nikon A1R, Zeiss LSM 980 Airyscan) on an inverted microscope with an environmental enclosure. Use a resonant scanner for fast (30 fps) imaging. Set pinhole to 1 Airy unit for best sectioning or open slightly (1.2 AU) for more signal. Use 40-60x water-immersion objectives for live cells to match RI of aqueous media.

Common Reconstruction Algorithms

Airyscan joint deconvolution (Zeiss)
Richardson-Lucy with measured confocal PSF
Sparse deconvolution (Hessian regularization)
Deep-learning denoising (Noise2Fast, DnCNN)
Pixel reassignment (ISM) for resolution doubling

Common Mistakes

Setting pinhole too small, drastically reducing signal in live cells
Scanning too slowly, causing phototoxicity and photobleaching
Using oil-immersion objectives for aqueous samples, introducing spherical aberration
Ignoring chromatic aberration when imaging multiple channels simultaneously
Oversampling (too many pixels) leading to excessive total dose with no resolution gain

How to Avoid Mistakes

Match pinhole to 1 AU and use resonant scanning + frame averaging for speed
Minimize pixel dwell time and total exposure; use sensitive GaAsP detectors
Select water-immersion objectives for live aqueous samples
Calibrate chromatic offsets with multi-color beads and apply corrections
Follow Nyquist sampling (pixel size ~ 0.4× resolution limit); avoid oversampling

Forward-Model Mismatch Cases

The widefield fallback uses sigma=2.0, but confocal PSF is sharper (sigma~1.2-1.5) due to the pinhole rejecting out-of-focus light — the fallback over-blurs by 30-60%, destroying resolvable features
Confocal provides optical sectioning (only in-focus plane contributes signal), while widefield collects fluorescence from all planes — reconstructions using widefield PSF will have incorrect out-of-focus model

How to Correct the Mismatch

Use the confocal operator with the correct PSF (product of excitation and detection PSFs, effective sigma~1.2-1.5) matching the pinhole size and objective NA
Model the confocal sectioning effect explicitly; for live-cell work, use the confocal PSF that accounts for pinhole size (1 Airy unit) and emission wavelength

Experimental Setup — Signal Chain

Experimental setup diagram for Confocal Live-Cell Microscopy

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

17.022206064497794 dB SSIM 0.38380508849556166

II — Mismatch

Ground Truth

Measurement

Reconstruction

14.93801985037615 dB SSIM 0.24399594480247574

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

20.039527108294998 dB SSIM 0.5012899560940253

I — Ideal

Ground Truth

Measurement

Reconstruction

14.231375949985203 dB SSIM 0.29689664293183293

II — Mismatch

Ground Truth

Measurement

Reconstruction

12.92413737908083 dB SSIM 0.20980780569697918

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.225830712320835 dB SSIM 0.547962131323724

I — Ideal

Ground Truth

Measurement

Reconstruction

8.520061941863302 dB SSIM 0.3871062840596513

II — Mismatch

Ground Truth

Measurement

Reconstruction

8.013928482144143 dB SSIM 0.23077294300010404

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

20.113884181775745 dB SSIM 0.3416888134023018

I — Ideal

Ground Truth

Measurement

Reconstruction

13.297073801541249 dB SSIM 0.5284335563352818

II — Mismatch

Ground Truth

Measurement

Reconstruction

11.66205075678512 dB SSIM 0.26941399804629346

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.61468818224941 dB SSIM 0.44080752632935544

Mean PSNR Across All Scenes

13.267679439471888 dB

I — Ideal

11.884534117096562 dB

II — Mismatch

19.748482546160247 dB

III — Oracle

Degradation: -1.4 dB Recovery: +7.9 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	17.022206064497794	0.38380508849556166	14.93801985037615	0.24399594480247574	20.039527108294998	0.5012899560940253
scene_01	14.231375949985203	0.29689664293183293	12.92413737908083	0.20980780569697918	19.225830712320835	0.547962131323724
scene_02	8.520061941863302	0.3871062840596513	8.013928482144143	0.23077294300010404	20.113884181775745	0.3416888134023018
scene_03	13.297073801541249	0.5284335563352818	11.66205075678512	0.26941399804629346	19.61468818224941	0.44080752632935544
Mean	13.267679439471888	0.39906039295558193	11.884534117096562	0.23849767288646312	19.748482546160247	0.45793710678735156

Experimental Setup

Instrument: Zeiss LSM 880 / Nikon A1R HD25

Objective: Plan Apo 63x / 1.40 NA oil

Pixel Size Nm: 80

Excitation Source: 488 nm Argon laser (5 mW)

Pinhole Au: 1.0

Dwell Time Us: 2

Frame Interval S: 5

Time Points: 200

Image Size: 512x512

Detector: GaAsP PMT (Zeiss Airyscan or Nikon spectral detector)

Key References

Minsky, 'Memoir on inventing the confocal microscope', Scanning 10, 128-138 (1988)
McNally et al., 'Three-dimensional imaging by deconvolution microscopy', Methods 23, 210-217 (1999)

Canonical Datasets

Cell Tracking Challenge confocal sequences
BioSR confocal subset

Spec DAG — Forward Model Pipeline

C(PSF_confocal) → D(g, η₃)

C Confocal PSF (PSF_confocal)

D PMT / HyD (g, η₃)

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
Δph	pinhole	Pinhole diameter error (μm)	0	5.0
Δn	refractive_index	Refractive index mismatch	1.515	1.52
α_b	photobleaching	Photobleaching rate error (%)	0	5.0

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.