PALM/STORM

PALM/STORM Single-Molecule Localization

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
🥇 DECODE 0.743 32.1 0.915 ✓ Certified Speiser et al., Nat. Methods 2021
🥈 Deep-STORM 0.693 30.2 0.880 ✓ Certified Nehme et al., Optica 2018
🥉 FALCON 0.550 25.8 0.740 ✓ Certified Min et al., Sci. Rep. 2014
4 ThunderSTORM 0.430 22.5 0.610 ✓ Certified Ovesny et al., Bioinformatics 2014

Dataset: PWM Benchmark (4 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
🥇 DECODE + gradient 0.674
0.741
30.01 dB / 0.917
0.672
26.52 dB / 0.846
0.608
23.55 dB / 0.752
✓ Certified Speiser et al., Nat. Methods 2021
🥈 Deep-STORM + gradient 0.587
0.737
29.17 dB / 0.903
0.545
21.6 dB / 0.673
0.478
18.86 dB / 0.543
✓ Certified Nehme et al., Optica 2018
🥉 FALCON + gradient 0.540
0.641
24.21 dB / 0.776
0.506
19.82 dB / 0.590
0.474
19.21 dB / 0.560
✓ Certified Min et al., Sci. Rep. 2014
4 ThunderSTORM + gradient 0.460
0.522
20.24 dB / 0.610
0.451
17.9 dB / 0.495
0.408
17.19 dB / 0.460
✓ Certified Ovesny et al., Bioinformatics 2014

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 DECODE + gradient 0.741 30.01 0.917
2 Deep-STORM + gradient 0.737 29.17 0.903
3 FALCON + gradient 0.641 24.21 0.776
4 ThunderSTORM + gradient 0.522 20.24 0.61
Spec Ranges (3 parameters)
Parameter Min Max Unit
psf_model -5.0 10.0 GaussianvsAiry
emitter_density -20.0 40.0 %
drift -0.5 1.0 nm/frame
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 DECODE + gradient 0.672 26.52 0.846
2 Deep-STORM + gradient 0.545 21.6 0.673
3 FALCON + gradient 0.506 19.82 0.59
4 ThunderSTORM + gradient 0.451 17.9 0.495
Spec Ranges (3 parameters)
Parameter Min Max Unit
psf_model -6.0 9.0 GaussianvsAiry
emitter_density -24.0 36.0 %
drift -0.6 0.9 nm/frame
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 DECODE + gradient 0.608 23.55 0.752
2 Deep-STORM + gradient 0.478 18.86 0.543
3 FALCON + gradient 0.474 19.21 0.56
4 ThunderSTORM + gradient 0.408 17.19 0.46
Spec Ranges (3 parameters)
Parameter Min Max Unit
psf_model -3.5 11.5 GaussianvsAiry
emitter_density -14.0 46.0 %
drift -0.35 1.15 nm/frame
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

Photoactivated localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM) achieve nanoscale resolution by stochastically activating sparse subsets of fluorescent molecules per frame, localizing each with sub-diffraction precision (proportional to sigma/sqrt(N) where N is detected photons), and accumulating localizations over thousands of frames. Typical localization precision is 10-30 nm. Primary challenges include overlapping emitters at high density, sample drift, and blinking statistics. Reconstruction uses Gaussian fitting (ThunderSTORM) or deep learning (DECODE).

Principle

Single-Molecule Localization Microscopy (PALM/STORM) achieves ~20 nm resolution by stochastically switching individual fluorophores between bright and dark states. In each frame, only a sparse subset of molecules emit, allowing their positions to be localized with sub-pixel precision by fitting 2-D Gaussians. Thousands of frames are accumulated and all localizations are plotted to form a super-resolution image.

How to Build the System

Use a TIRF microscope (100x 1.49 NA oil objective) with powerful laser excitation (200-500 mW at the sample, 647 nm for Alexa647 STORM or 561 nm for mEos PALM). TIRF geometry reduces background. An oxygen-scavenging buffer with thiol (MEA/BME) is critical for Alexa647 blinking. Use an EMCCD (Andor iXon 897) or fast sCMOS camera at 30-100 Hz frame rate. Acquire 10,000-50,000 frames.

Common Reconstruction Algorithms

  • ThunderSTORM (ImageJ plugin, MLE/LSQ Gaussian fitting)
  • SMLM ZOLA-3D (deep-learning 3D localization)
  • DAOSTORM (multi-emitter fitting for high density)
  • Drift correction (fiducial-based or cross-correlation)
  • HAWK / ANNA-PALM (deep-learning for accelerated SMLM)

Common Mistakes

  • Density of active emitters too high, causing overlapping PSFs and localization errors
  • Insufficient photon count per localization, yielding poor precision (>30 nm)
  • Sample drift during long acquisitions not corrected
  • Poor blinking statistics (incomplete on-off switching) from wrong buffer conditions
  • Mistaking fixed-pattern noise or autofluorescence for single molecules

How to Avoid Mistakes

  • Tune activation laser to achieve sparse single-molecule density per frame
  • Optimize buffer (pH, thiol concentration, oxygen scavenger) for bright blinks (>1000 photons)
  • Include fiducial markers (gold beads or TetraSpeck) and apply drift correction
  • Prepare fresh imaging buffer immediately before acquisition; degas thoroughly
  • Apply quality filters (photon threshold, localization precision, PSF shape) in analysis

Forward-Model Mismatch Cases

  • The widefield fallback produces a blurred intensity image, but PALM/STORM generates sparse single-molecule localizations — the correct forward model produces a list of (x,y,photons) events, not a convolved image
  • Using a continuous PSF blur instead of the discrete point-emitter model (y = sum_i(n_i * PSF(r - r_i) + background)) means single-molecule fitting algorithms will receive incorrect input and localization precision estimates will be meaningless

How to Correct the Mismatch

  • Use the PALM/STORM operator that simulates stochastic single-molecule activation: sparse emitters with Poisson photon counts, individually convolved with the PSF, on a per-frame basis
  • Reconstruct using single-molecule localization (Gaussian fitting, MLE) on the correct sparse-emitter frames; the forward model must match the blinking kinetics and photon statistics of the fluorophore

Experimental Setup — Signal Chain

Experimental setup diagram for PALM/STORM Single-Molecule Localization

Experimental Setup

Instrument: Nikon N-STORM / Zeiss ELYRA 7 SMLM
Objective: Apo TIRF 100x / 1.49 NA oil
Camera Pixel Nm: 100
Reconstruction Pixel Nm: 25
Excitation Source: 640 nm laser (200 mW at fiber tip)
Activation Laser Nm: 405
Exposure Ms: 20
Total Frames: 10000
Frame Rate Fps: 50
Achieved Resolution Nm: 20
Detector: Andor iXon Ultra 897 EMCCD
Imaging Buffer: MEA + GLOX for Alexa Fluor 647

Key References

  • Betzig et al., 'Imaging intracellular fluorescent proteins at nanometer resolution', Science 313, 1642-1645 (2006)
  • Rust et al., 'Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM)', Nature Methods 3, 793-796 (2006)
  • Speiser et al., 'Deep learning enables fast and dense single-molecule localization (DECODE)', Nature Methods 18, 1082-1090 (2021)

Canonical Datasets

  • SMLM Challenge 2016 (Sage et al., Nature Methods 2019)
  • ThunderSTORM tutorial datasets

Spec DAG — Forward Model Pipeline

C(PSF) → D(g, η₃)

C Single-Molecule PSF (PSF)
D EMCCD / sCMOS (g, η₃)

Mismatch Parameters

Symbol Parameter Description Nominal Perturbed
ΔPSF psf_model PSF model error (Gaussian vs Airy) 0 5.0
Δρ emitter_density Emitter density error (%) 0 20.0
Δr drift Stage drift (nm/frame) 0 0.5

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.