FLIM

Fluorescence Lifetime Imaging

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
🥇	DiffFLIM DiffFLIM Gao et al. 2024 39.6 dB SSIM 0.957 Checkpoint unavailable	0.889	39.6	0.957	✓ Certified	Gao et al. 2024
🥈	PhysFLIM PhysFLIM Chen et al. 2024 38.2 dB SSIM 0.945 Checkpoint unavailable	0.859	38.2	0.945	✓ Certified	Chen et al. 2024
🥉	SwinFLIM SwinFLIM Zhang et al. 2023 37.0 dB SSIM 0.935 Checkpoint unavailable	0.834	37.0	0.935	✓ Certified	Zhang et al. 2023
4	TransFLIM TransFLIM Wang et al. 2022 35.5 dB SSIM 0.918 Checkpoint unavailable	0.801	35.5	0.918	✓ Certified	Wang et al. 2022
5	FLIMJ FLIMJ Li et al. 2022 33.1 dB SSIM 0.882 Checkpoint unavailable	0.743	33.1	0.882	✓ Certified	Li et al. 2022
6	DnCNN-FLIM DnCNN-FLIM Smith et al. 2019 30.7 dB SSIM 0.845 Checkpoint unavailable	0.684	30.7	0.845	✓ Certified	Smith et al. 2019
7	RLD-FLIM RLD-FLIM Ballew & Demas 1989 27.9 dB SSIM 0.798 Try in SpecLab →	0.614	27.9	0.798	✓ Certified	Ballew & Demas 1989
8	MLE-FLIM MLE-FLIM Grinvald & Steinberg 1974 25.8 dB SSIM 0.762 Try in SpecLab →	0.561	25.8	0.762	✓ Certified	Grinvald & Steinberg 1974
9	Phasor-FLIM Phasor-FLIM Digman et al. 2008 23.2 dB SSIM 0.722 Try in SpecLab →	0.498	23.2	0.722	✓ Certified	Digman et al. 2008

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
🥇	SwinFLIM + gradient SwinFLIM + gradient Zhang et al., Biomed. Opt. Express 2023 Score 0.774 Correct & Reconstruct →	0.774	0.807 34.04 dB / 0.961	0.775 33.16 dB / 0.954	0.741 31.31 dB / 0.935	✓ Certified	Zhang et al., Biomed. Opt. Express 2023
🥈	PhysFLIM + gradient PhysFLIM + gradient Chen et al., Nat. Photonics 2024 Score 0.758 Correct & Reconstruct →	0.758	0.825 36.17 dB / 0.974	0.745 30.41 dB / 0.923	0.703 29.04 dB / 0.901	✓ Certified	Chen et al., Nat. Photonics 2024
🥉	DiffFLIM + gradient DiffFLIM + gradient Gao et al., NeurIPS 2024 Score 0.751 Correct & Reconstruct →	0.751	0.842 37.05 dB / 0.978	0.727 29.06 dB / 0.901	0.685 27.71 dB / 0.875	✓ Certified	Gao et al., NeurIPS 2024
4	TransFLIM + gradient TransFLIM + gradient Wang et al., Nat. Methods 2022 Score 0.716 Correct & Reconstruct →	0.716	0.787 32.53 dB / 0.948	0.728 29.72 dB / 0.912	0.634 24.79 dB / 0.795	✓ Certified	Wang et al., Nat. Methods 2022
5	FLIMJ + gradient FLIMJ + gradient Li et al., Nat. Methods 2022 Score 0.674 Correct & Reconstruct →	0.674	0.758 30.97 dB / 0.930	0.650 25.15 dB / 0.807	0.614 23.76 dB / 0.760	✓ Certified	Li et al., Nat. Methods 2022
6	RLD-FLIM + gradient RLD-FLIM + gradient Ballew & Demas, Anal. Chem. 1989 Score 0.656 Correct & Reconstruct →	0.656	0.686 26.14 dB / 0.836	0.647 25.72 dB / 0.824	0.635 25.2 dB / 0.808	✓ Certified	Ballew & Demas, Anal. Chem. 1989
7	DnCNN-FLIM + gradient DnCNN-FLIM + gradient Smith et al., Nat. Methods 2019 Score 0.587 Correct & Reconstruct →	0.587	0.718 28.68 dB / 0.894	0.570 22.35 dB / 0.705	0.473 19.25 dB / 0.562	✓ Certified	Smith et al., Nat. Methods 2019
8	Phasor-FLIM + gradient Phasor-FLIM + gradient Digman et al., Biophys. J. 2008 Score 0.502 Correct & Reconstruct →	0.502	0.575 21.66 dB / 0.675	0.481 19.37 dB / 0.568	0.450 18.41 dB / 0.520	✓ Certified	Digman et al., Biophys. J. 2008
9	MLE-FLIM + gradient MLE-FLIM + gradient Grinvald & Steinberg, Anal. Biochem. 1974 Score 0.433 Correct & Reconstruct →	0.433	0.600 22.82 dB / 0.724	0.391 15.87 dB / 0.395	0.309 13.8 dB / 0.302	✓ Certified	Grinvald & Steinberg, Anal. Biochem. 1974

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	DiffFLIM + gradient	0.842	37.05	0.978
2	PhysFLIM + gradient	0.825	36.17	0.974
3	SwinFLIM + gradient	0.807	34.04	0.961
4	TransFLIM + gradient	0.787	32.53	0.948
5	FLIMJ + gradient	0.758	30.97	0.93
6	DnCNN-FLIM + gradient	0.718	28.68	0.894
7	RLD-FLIM + gradient	0.686	26.14	0.836
8	MLE-FLIM + gradient	0.600	22.82	0.724
9	Phasor-FLIM + gradient	0.575	21.66	0.675

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
irf_width	-20.0	40.0	ps
time_bin	-5.0	10.0	ps
afterpulsing	-0.005	0.01

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	SwinFLIM + gradient	0.775	33.16	0.954
2	PhysFLIM + gradient	0.745	30.41	0.923
3	TransFLIM + gradient	0.728	29.72	0.912
4	DiffFLIM + gradient	0.727	29.06	0.901
5	FLIMJ + gradient	0.650	25.15	0.807
6	RLD-FLIM + gradient	0.647	25.72	0.824
7	DnCNN-FLIM + gradient	0.570	22.35	0.705
8	Phasor-FLIM + gradient	0.481	19.37	0.568
9	MLE-FLIM + gradient	0.391	15.87	0.395

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
irf_width	-24.0	36.0	ps
time_bin	-6.0	9.0	ps
afterpulsing	-0.006	0.009

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	SwinFLIM + gradient	0.741	31.31	0.935
2	PhysFLIM + gradient	0.703	29.04	0.901
3	DiffFLIM + gradient	0.685	27.71	0.875
4	RLD-FLIM + gradient	0.635	25.2	0.808
5	TransFLIM + gradient	0.634	24.79	0.795
6	FLIMJ + gradient	0.614	23.76	0.76
7	DnCNN-FLIM + gradient	0.473	19.25	0.562
8	Phasor-FLIM + gradient	0.450	18.41	0.52
9	MLE-FLIM + gradient	0.309	13.8	0.302

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
irf_width	-14.0	46.0	ps
time_bin	-3.5	11.5	ps
afterpulsing	-0.0035	0.0115

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

Fluorescence lifetime imaging microscopy (FLIM) measures the exponential decay time of fluorescence emission at each pixel, providing contrast based on the molecular environment rather than intensity alone. In time-correlated single-photon counting (TCSPC), each detected photon is time-tagged relative to the excitation pulse, building a histogram of arrival times that is fitted to single- or multi-exponential decay models. The phasor approach provides a fit-free analysis in Fourier space. Primary challenges include low photon counts and instrument response function (IRF) deconvolution.

Principle

Fluorescence Lifetime Imaging measures the exponential decay time of fluorophore emission (typically 1-10 ns) rather than intensity. Lifetime is sensitive to the fluorophore's local chemical environment (pH, ion concentration, FRET) but independent of concentration and photobleaching. Detection uses either time-correlated single-photon counting (TCSPC) or frequency-domain phase/modulation methods.

How to Build the System

Add a pulsed laser source (ps diode laser or Ti:Sapphire, 40-80 MHz repetition rate) to a confocal or widefield microscope. For TCSPC, install single-photon counting detectors (hybrid PMTs or SPADs) with timing electronics (Becker & Hickl SPC-150/830 or PicoQuant TimeHarp). For widefield FLIM, use a gated or modulated camera (Lambert Instruments). Synchronize laser pulses with detector timing.

Common Reconstruction Algorithms

Mono-exponential / bi-exponential tail fitting (least-squares or MLE)
Phasor analysis (model-free lifetime decomposition)
Global analysis (linked lifetime fitting across pixels)
Bayesian lifetime estimation
Deep-learning FLIM (FLIMnet, rapid lifetime prediction from few photons)

Common Mistakes

Insufficient photon counts for reliable lifetime fitting (need ≥1000 photons/pixel)
Ignoring instrument response function (IRF) convolution in the fit
Using mono-exponential fit for multi-component decays, obtaining incorrect average lifetimes
Pile-up effect at high count rates distorting the decay histogram
Background autofluorescence contributing a long-lifetime component

How to Avoid Mistakes

Collect sufficient photons; use longer acquisition or binning if needed
Measure IRF with a scattering sample and convolve with the model in fitting
Evaluate fit residuals; use bi-exponential or phasor if mono-exponential is poor
Keep count rate below 1-5 % of the laser repetition rate to avoid pile-up
Measure autofluorescence lifetime separately and include in the fit model

Forward-Model Mismatch Cases

The widefield fallback produces a single 2D intensity image (64,64), but FLIM measures fluorescence lifetime decay at each pixel — output shape (64,64,64) includes the temporal decay dimension
FLIM forward model is nonlinear (exponential decay convolved with IRF: y(t) = IRF * sum(a_i * exp(-t/tau_i))), while the widefield linear blur cannot represent lifetime information at all

How to Correct the Mismatch

Use the FLIM operator that generates time-resolved fluorescence decay histograms at each pixel, including IRF convolution and multi-exponential decay components
Reconstruct lifetimes using phasor analysis or exponential fitting on the temporal dimension; the correct forward model preserves the relationship between decay time and local chemical environment

Experimental Setup — Signal Chain

Experimental setup diagram for Fluorescence Lifetime Imaging

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: Becker & Hickl SPC-150N with Zeiss LSM 880

Objective: Plan Apo 63x / 1.30 NA oil

Pixel Size Nm: 100

Image Size: 256x256

Tcspc Channels: 256

Time Resolution Ps: 50

Irf Fwhm Ps: 25

Excitation Source: pulsed diode laser (405 nm, 40 MHz repetition)

Repetition Rate Mhz: 40

Lifetime Range Ns: 0.5-10

Detector: Hybrid PMT (Becker & Hickl HPM-100-40)

Analysis: phasor / bi-exponential fit

Key References

Becker, 'Advanced Time-Correlated Single Photon Counting Techniques', Springer (2005)
Digman et al., 'The phasor approach to fluorescence lifetime imaging', Biophysical Journal 94, L14-L16 (2008)

Canonical Datasets

FLIM-FRET standard sample datasets (Becker & Hickl)
FLIM phasor benchmark (Digman lab)

Spec DAG — Forward Model Pipeline

C(PSF) → Σ_t → D(g, η₃)

C PSF Convolution (PSF)

Σ Time-Gate Sum (t)

D TCSPC Detector (g, η₃)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
ΔIRF	irf_width	Instrument response width error (ps)	20
Δτ_b	time_bin	Time bin error (ps)	5
p_ap	afterpulsing	Afterpulsing probability	0.005

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.