OCTA

OCT Angiography

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
🥇	ScoreOCT ScoreOCT Wei et al., ECCV 2025 37.95 dB SSIM 0.973 Checkpoint unavailable	0.869	37.95	0.973	✓ Certified	Wei et al., ECCV 2025
🥈	DiffusionOCT DiffusionOCT Zhang et al., NeurIPS 2024 37.52 dB SSIM 0.970 Checkpoint unavailable	0.860	37.52	0.970	✓ Certified	Zhang et al., NeurIPS 2024
🥉	SpeckleFormer SpeckleFormer Devalla et al., ECCV 2024 36.85 dB SSIM 0.964 Checkpoint unavailable	0.846	36.85	0.964	✓ Certified	Devalla et al., ECCV 2024
4	RetinalFormer RetinalFormer Chen et al., ICCV 2024 36.35 dB SSIM 0.960 Checkpoint unavailable	0.836	36.35	0.960	✓ Certified	Chen et al., ICCV 2024
5	OCT-ViT OCT-ViT Tian et al., ICCV 2024 36.12 dB SSIM 0.958 Checkpoint unavailable	0.831	36.12	0.958	✓ Certified	Tian et al., ICCV 2024
6	OCTA-Net OCTA-Net Hybrid U-Net+Transformer, 2023 34.6 dB SSIM 0.942 Checkpoint unavailable	0.798	34.6	0.942	✓ Certified	Hybrid U-Net+Transformer, 2023
7	U-Net-OCT U-Net-OCT U-Net variant 33.85 dB SSIM 0.935 Checkpoint unavailable	0.782	33.85	0.935	✓ Certified	U-Net variant
8	Speckle-DenoiseNet Speckle-DenoiseNet Devalla et al., BOE 2019 33.1 dB SSIM 0.925 Checkpoint unavailable	0.764	33.1	0.925	✓ Certified	Devalla et al., BOE 2019
9	NLM-OCT NLM-OCT Non-local means variant 30.2 dB SSIM 0.870 Try in SpecLab →	0.688	30.2	0.870	✓ Certified	Non-local means variant
10	BM4D BM4D Maggioni et al., IEEE TIP 2013 29.3 dB SSIM 0.850 Try in SpecLab →	0.663	29.3	0.850	✓ Certified	Maggioni et al., IEEE TIP 2013
11	TV-Denoising TV-Denoising TV regularization 28.5 dB SSIM 0.815 Try in SpecLab →	0.632	28.5	0.815	✓ Certified	TV regularization
12	Speckle-Lee Speckle-Lee Lee, IEEE TGRS 1980 27.85 dB SSIM 0.790 Try in SpecLab →	0.609	27.85	0.790	✓ Certified	Lee, IEEE TGRS 1980
13	FFT-OCT FFT-OCT Analytical baseline 25.6 dB SSIM 0.720 Try in SpecLab →	0.537	25.6	0.720	✓ Certified	Analytical baseline

Dataset: PWM Benchmark (13 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
🥇	DiffusionOCT + gradient DiffusionOCT + gradient Zhang et al., NeurIPS 2024 Score 0.776 Correct & Reconstruct →	0.776	0.838 36.2 dB / 0.974	0.766 31.91 dB / 0.942	0.725 30.14 dB / 0.919	✓ Certified	Zhang et al., NeurIPS 2024
🥈	OCT-ViT + gradient OCT-ViT + gradient Tian et al., ICCV 2024 Score 0.767 Correct & Reconstruct →	0.767	0.799 33.73 dB / 0.959	0.762 32.18 dB / 0.945	0.739 30.66 dB / 0.926	✓ Certified	Tian et al., ICCV 2024
🥉	ScoreOCT + gradient ScoreOCT + gradient Wei et al., ECCV 2025 Score 0.761 Correct & Reconstruct →	0.761	0.820 35.19 dB / 0.969	0.749 31.04 dB / 0.931	0.715 28.35 dB / 0.888	✓ Certified	Wei et al., ECCV 2025
4	SpeckleFormer + gradient SpeckleFormer + gradient Devalla et al., ECCV 2024 Score 0.744 Correct & Reconstruct →	0.744	0.831 35.81 dB / 0.972	0.750 30.95 dB / 0.930	0.652 25.68 dB / 0.823	✓ Certified	Devalla et al., ECCV 2024
5	RetinalFormer + gradient RetinalFormer + gradient Chen et al., ICCV 2024 Score 0.718 Correct & Reconstruct →	0.718	0.802 33.8 dB / 0.959	0.722 28.63 dB / 0.893	0.629 24.48 dB / 0.785	✓ Certified	Chen et al., ICCV 2024
6	OCTA-Net + gradient OCTA-Net + gradient Hybrid U-Net+Transformer, 2023 Score 0.715 Correct & Reconstruct →	0.715	0.803 33.6 dB / 0.958	0.695 28.03 dB / 0.881	0.646 25.97 dB / 0.831	✓ Certified	Hybrid U-Net+Transformer, 2023
7	U-Net-OCT + gradient U-Net-OCT + gradient Ronneberger et al., MICCAI 2015 (OCT variant) Score 0.666 Correct & Reconstruct →	0.666	0.789 32.45 dB / 0.947	0.650 25.41 dB / 0.815	0.559 22.12 dB / 0.695	✓ Certified	Ronneberger et al., MICCAI 2015 (OCT variant)
8	NLM-OCT + gradient NLM-OCT + gradient Buades et al., Multiscale Model. Simul. 2005 Score 0.663 Correct & Reconstruct →	0.663	0.708 28.11 dB / 0.883	0.673 26.6 dB / 0.848	0.608 24.35 dB / 0.781	✓ Certified	Buades et al., Multiscale Model. Simul. 2005
9	TV-Denoising + gradient TV-Denoising + gradient Rudin et al., Phys. A 1992 Score 0.647 Correct & Reconstruct →	0.647	0.673 26.22 dB / 0.838	0.660 26.15 dB / 0.836	0.608 24.52 dB / 0.787	✓ Certified	Rudin et al., Phys. A 1992
10	BM4D + gradient BM4D + gradient Maggioni et al., IEEE TIP 2013 Score 0.633 Correct & Reconstruct →	0.633	0.715 27.61 dB / 0.872	0.606 24.1 dB / 0.772	0.578 22.81 dB / 0.724	✓ Certified	Maggioni et al., IEEE TIP 2013
11	Speckle-DenoiseNet + gradient Speckle-DenoiseNet + gradient Devalla et al., BOE 2019 Score 0.623 Correct & Reconstruct →	0.623	0.758 31.23 dB / 0.934	0.610 23.85 dB / 0.763	0.500 20.43 dB / 0.619	✓ Certified	Devalla et al., BOE 2019
12	Speckle-Lee + gradient Speckle-Lee + gradient Lee, IEEE TGRS 1980 Score 0.617 Correct & Reconstruct →	0.617	0.654 25.12 dB / 0.806	0.623 24.47 dB / 0.785	0.575 22.2 dB / 0.698	✓ Certified	Lee, IEEE TGRS 1980
13	FFT-OCT + gradient FFT-OCT + gradient Analytical baseline Score 0.566 Correct & Reconstruct →	0.566	0.612 23.63 dB / 0.755	0.558 22.06 dB / 0.693	0.527 21.1 dB / 0.650	✓ Certified	Analytical baseline

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	DiffusionOCT + gradient	0.838	36.2	0.974
2	SpeckleFormer + gradient	0.831	35.81	0.972
3	ScoreOCT + gradient	0.820	35.19	0.969
4	OCTA-Net + gradient	0.803	33.6	0.958
5	RetinalFormer + gradient	0.802	33.8	0.959
6	OCT-ViT + gradient	0.799	33.73	0.959
7	U-Net-OCT + gradient	0.789	32.45	0.947
8	Speckle-DenoiseNet + gradient	0.758	31.23	0.934
9	BM4D + gradient	0.715	27.61	0.872
10	NLM-OCT + gradient	0.708	28.11	0.883
11	TV-Denoising + gradient	0.673	26.22	0.838
12	Speckle-Lee + gradient	0.654	25.12	0.806
13	FFT-OCT + gradient	0.612	23.63	0.755

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
inter_bscan_time	-0.5	1.0	ms
bulk_motion	-0.2	0.4	mm/s
decorrelation_threshold	0.45	0.6

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	DiffusionOCT + gradient	0.766	31.91	0.942
2	OCT-ViT + gradient	0.762	32.18	0.945
3	SpeckleFormer + gradient	0.750	30.95	0.93
4	ScoreOCT + gradient	0.749	31.04	0.931
5	RetinalFormer + gradient	0.722	28.63	0.893
6	OCTA-Net + gradient	0.695	28.03	0.881
7	NLM-OCT + gradient	0.673	26.6	0.848
8	TV-Denoising + gradient	0.660	26.15	0.836
9	U-Net-OCT + gradient	0.650	25.41	0.815
10	Speckle-Lee + gradient	0.623	24.47	0.785
11	Speckle-DenoiseNet + gradient	0.610	23.85	0.763
12	BM4D + gradient	0.606	24.1	0.772
13	FFT-OCT + gradient	0.558	22.06	0.693

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
inter_bscan_time	-0.6	0.9	ms
bulk_motion	-0.24	0.36	mm/s
decorrelation_threshold	0.44	0.59

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	OCT-ViT + gradient	0.739	30.66	0.926
2	DiffusionOCT + gradient	0.725	30.14	0.919
3	ScoreOCT + gradient	0.715	28.35	0.888
4	SpeckleFormer + gradient	0.652	25.68	0.823
5	OCTA-Net + gradient	0.646	25.97	0.831
6	RetinalFormer + gradient	0.629	24.48	0.785
7	NLM-OCT + gradient	0.608	24.35	0.781
8	TV-Denoising + gradient	0.608	24.52	0.787
9	BM4D + gradient	0.578	22.81	0.724
10	Speckle-Lee + gradient	0.575	22.2	0.698
11	U-Net-OCT + gradient	0.559	22.12	0.695
12	FFT-OCT + gradient	0.527	21.1	0.65
13	Speckle-DenoiseNet + gradient	0.500	20.43	0.619

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
inter_bscan_time	-0.35	1.15	ms
bulk_motion	-0.14	0.46	mm/s
decorrelation_threshold	0.465	0.615

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

OCT angiography extends standard OCT by acquiring repeated B-scans at the same location and computing the decorrelation of the complex OCT signal between successive scans. Moving red blood cells cause temporal fluctuations that differ from static tissue, enabling label-free visualization of retinal vasculature. The contrast mechanism uses amplitude decorrelation (SSADA), phase variance, or complex-signal algorithms. Key limitations include motion artifacts, projection artifacts from superficial vessels, and limited field of view.

Principle

OCT Angiography detects blood flow non-invasively by comparing repeated OCT B-scans at the same location. Moving red blood cells cause temporal fluctuations in the OCT signal (amplitude and/or phase), while static tissue remains constant. Decorrelation, variance, or differential analysis between repeated scans produces a motion-contrast image revealing the vasculature without the need for injectable contrast agents.

How to Build the System

Use a high-speed OCT system (≥70 kHz A-scan rate, swept-source preferred) capable of repeated B-scans at the same location. Acquire 2-4 repeated B-scans at each position with inter-scan time of 3-10 ms. An eye-tracking system is essential for ophthalmic OCTA to correct microsaccades. Process with split-spectrum amplitude-decorrelation (SSADA), optical microangiography (OMAG), or phase-variance algorithms.

Common Reconstruction Algorithms

SSADA (Split-Spectrum Amplitude-Decorrelation Angiography)
OMAG (Optical Micro-Angiography, complex signal differential)
Phase-variance OCTA
Deep-learning OCTA denoising and vessel segmentation
Projection artifact removal algorithms

Common Mistakes

Bulk tissue motion producing decorrelation artifacts (false flow signals)
Projection artifacts where superficial vessel shadows appear in deeper layers
Shadow artifacts beneath large vessels causing false flow voids
Insufficient inter-scan interval for detecting slow capillary flow
Motion artifacts from blinks or microsaccades corrupting OCTA volumes

How to Avoid Mistakes

Apply bulk motion correction (axial and lateral registration) before decorrelation analysis
Use projection artifact removal algorithms (slab subtraction or OMAG-based)
Increase number of repeated B-scans to improve SNR and reduce shadow impact
Optimize inter-scan time: shorter for fast flow, longer for slow capillary flow
Use active eye tracking and discard frames with large motion; average multiple volumes

Forward-Model Mismatch Cases

The widefield fallback applies static spatial blur, but OCTA detects blood flow by comparing repeated OCT B-scans — the temporal decorrelation between scans caused by moving red blood cells is not modeled
OCTA is fundamentally a motion-contrast technique (flow signal = decorrelation or variance between repeated measurements) — the widefield static model has no temporal dimension and cannot detect or distinguish flowing from static tissue

How to Correct the Mismatch

Use the OCTA operator that models repeated OCT measurements at the same location: static tissue produces correlated signals while flowing blood produces decorrelated signals between repeated scans
Extract flow maps using SSADA (split-spectrum amplitude decorrelation) or OMAG (optical microangiography) that require multiple temporally separated OCT measurements as input

Experimental Setup — Signal Chain

Experimental setup diagram for OCT Angiography

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

21.7701889386771 dB SSIM 0.28157968449959064

II — Mismatch

Ground Truth

Measurement

Reconstruction

17.221639470926625 dB SSIM 0.06789440210933223

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

18.279654695337403 dB SSIM 0.12832852712757956

I — Ideal

Ground Truth

Measurement

Reconstruction

20.94482753471155 dB SSIM 0.2832431148687715

II — Mismatch

Ground Truth

Measurement

Reconstruction

16.981433781163226 dB SSIM 0.07672767180609508

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.75430085731928 dB SSIM 0.1808684012136474

I — Ideal

Ground Truth

Measurement

Reconstruction

21.61496340551778 dB SSIM 0.3159591496980528

II — Mismatch

Ground Truth

Measurement

Reconstruction

16.920769154324237 dB SSIM 0.07842711467674197

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

18.769205830799947 dB SSIM 0.18514709798791337

I — Ideal

Ground Truth

Measurement

Reconstruction

21.403159955514216 dB SSIM 0.2738592733230193

II — Mismatch

Ground Truth

Measurement

Reconstruction

16.233506684280766 dB SSIM 0.05907636169469176

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

17.993355974683976 dB SSIM 0.11967474616417093

Mean PSNR Across All Scenes

21.43328495860516 dB

I — Ideal

16.839337272673713 dB

II — Mismatch

18.69912933953515 dB

III — Oracle

Degradation: -4.6 dB Recovery: +1.9 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	21.7701889386771	0.28157968449959064	17.221639470926625	0.06789440210933223	18.279654695337403	0.12832852712757956
scene_01	20.94482753471155	0.2832431148687715	16.981433781163226	0.07672767180609508	19.75430085731928	0.1808684012136474
scene_02	21.61496340551778	0.3159591496980528	16.920769154324237	0.07842711467674197	18.769205830799947	0.18514709798791337
scene_03	21.403159955514216	0.2738592733230193	16.233506684280766	0.05907636169469176	17.993355974683976	0.11967474616417093
Mean	21.43328495860516	0.28866030559735856	16.839337272673713	0.07053138757171526	18.69912933953515	0.15350469312332782

Experimental Setup

Instrument: Zeiss PLEX Elite 9000 / Optovue AngioVue

Wavelength Nm: 840

A Scan Rate Khz: 68

Scan Pattern: 6x6 mm

Repeated B Scans: 4

En Face Resolution Um: 15

Algorithm: SSADA / OCTA ratio

Key References

Jia et al., 'Split-spectrum amplitude-decorrelation angiography (SSADA)', Opt. Express 20, 4710 (2012)
Spaide et al., 'OCT Angiography', Prog. Retin. Eye Res. 64, 1 (2018)

Canonical Datasets

OCTA-500 (Li et al., Scientific Data 2024)
ROSE retinal OCTA vessel segmentation

Spec DAG — Forward Model Pipeline

P(low-coherence) → Σ(interference) → D(g, η₁)

P Low-Coherence Source (low-coherence)

Σ Interferometric Sum (interference)

D Spectrometer (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Nominal	Perturbed
ΔT	inter_bscan_time	Inter-B-scan time error (ms)	0	0.5
Δv_b	bulk_motion	Bulk motion artifact (mm/s)	0	0.2
ΔD_th	decorrelation_threshold	Decorrelation threshold error	0.5	0.55

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.