MRI

Magnetic Resonance 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
🥇	SwinMR++ SwinMR++ Huang et al., IEEE TMI 2025 — 5 improvements: multi-scale axial attention (cross-scale long-range modeling), INR coordinate-query head (high-acceleration k-space interpolation), k-space DC per unrolled module, joint LPIPS+SSIM+k-space consistency loss, dynamic conv-Transformer branch weighting 43.8 dB SSIM 0.983 Checkpoint unavailable	0.971	43.8	0.983	✓ Certified	Huang et al., IEEE TMI 2025 — 5 improvements: multi-scale axial attention (cross-scale long-range modeling), INR coordinate-query head (high-acceleration k-space interpolation), k-space DC per unrolled module, joint LPIPS+SSIM+k-space consistency loss, dynamic conv-Transformer branch weighting
🥈	HUMUS-Net++ HUMUS-Net++ Fabian et al., dHUMUS-Net 2023 — 5 improvements: k-space DC per unrolled module, dynamic optimal-scale prediction (dHUMUS-Net), INR coordinate head (continuous representation), LPIPS+SSIM perceptual-structural loss, lightweight axial attention Transformer 43.1 dB SSIM 0.979 Checkpoint unavailable	0.958	43.1	0.979	✓ Certified	Fabian et al., dHUMUS-Net 2023 — 5 improvements: k-space DC per unrolled module, dynamic optimal-scale prediction (dHUMUS-Net), INR coordinate head (continuous representation), LPIPS+SSIM perceptual-structural loss, lightweight axial attention Transformer
🥉	MR-IPT MR-IPT Sci. Reports 2025 42.48 dB SSIM 0.983 Checkpoint unavailable	0.950	42.48	0.983	✓ Certified	Sci. Reports 2025
4	HybridCascade++ HybridCascade++ HybridCascade++ MICCAI 2021 + IEEE TMI 2025 — 5 improvements: multi-stage cascade DC (coarse-to-fine 4-stage unrolling), SIREN INR warm-start (continuous prior initialization), SSIM structural anchor (perceptual consistency in late DC stages), DRUNet final polish (blind denoising post-DC), freq-blend LF/HF fusion (SIREN low-freq + structured high-freq recombination) 42.5 dB SSIM 0.981 Checkpoint unavailable	0.949	42.5	0.981	✓ Certified	HybridCascade++ MICCAI 2021 + IEEE TMI 2025 — 5 improvements: multi-stage cascade DC (coarse-to-fine 4-stage unrolling), SIREN INR warm-start (continuous prior initialization), SSIM structural anchor (perceptual consistency in late DC stages), DRUNet final polish (blind denoising post-DC), freq-blend LF/HF fusion (SIREN low-freq + structured high-freq recombination)
5	MoDL-Net++ MoDL-Net++ MoDL-Net++ IEEE TMI 2025 — 5 improvements: multi-scale pyramid fusion (coarse-to-fine representation), RDN/Swin deep prior (rich feature hierarchy), differentiable DC layers (physics-informed unrolling), joint LPIPS+SSIM+L1 loss (perceptual+structural+fidelity), two-stage training (pre-train then fine-tune with DC) 41.8 dB SSIM 0.978 Checkpoint unavailable	0.936	41.8	0.978	✓ Certified	MoDL-Net++ IEEE TMI 2025 — 5 improvements: multi-scale pyramid fusion (coarse-to-fine representation), RDN/Swin deep prior (rich feature hierarchy), differentiable DC layers (physics-informed unrolling), joint LPIPS+SSIM+L1 loss (perceptual+structural+fidelity), two-stage training (pre-train then fine-tune with DC)
6	U-Net++ U-Net++ Chen & Boning, IEEE TMI 2024 — 5 improvements: Residual U-Net blocks (dense skip connections), data consistency layers (physics-informed k-space projection), plug-and-play prior (learned denoiser as proximal operator), joint SSIM+MSE+DC loss, multi-scale feature aggregation 41.5 dB SSIM 0.978 Checkpoint unavailable	0.931	41.5	0.978	✓ Certified	Chen & Boning, IEEE TMI 2024 — 5 improvements: Residual U-Net blocks (dense skip connections), data consistency layers (physics-informed k-space projection), plug-and-play prior (learned denoiser as proximal operator), joint SSIM+MSE+DC loss, multi-scale feature aggregation
7	MRI-FM MRI-FM Wang et al., Nature MI 2026 42.1 dB SSIM 0.948 Checkpoint unavailable	0.926	42.1	0.948	✓ Certified	Wang et al., Nature MI 2026
8	ReconFormer++ ReconFormer++ Pan et al., IEEE TMI 2025 41.5 dB SSIM 0.969 Checkpoint unavailable	0.926	41.5	0.969	✓ Certified	Pan et al., IEEE TMI 2025
9	PromptMR-SFM PromptMR-SFM PWM 2026 41.3 dB SSIM 0.971 Checkpoint unavailable	0.924	41.3	0.971	✓ Certified	PWM 2026
10	PnP-DnCNN-Pro PnP-DnCNN-Pro PnP-DnCNN-Pro IEEE TMI 2025 (DOI:10.1109/TMI.2025.3441240) — 5 improvements: multi-scale DnCNN denoiser (SwinIR-style hierarchical feature extraction), adaptive mu/sigma schedule (dynamic regularization per PnP iteration), SIREN INR coordinate output head (continuous representation for high-acceleration interpolation), joint LPIPS+SSIM denoiser training (perceptual+structural loss), dynamic PnP regularization scheduling (learnable lambda per iteration) 41.0 dB SSIM 0.968 Try in SpecLab →	0.917	41.0	0.968	✓ Certified	PnP-DnCNN-Pro IEEE TMI 2025 (DOI:10.1109/TMI.2025.3441240) — 5 improvements: multi-scale DnCNN denoiser (SwinIR-style hierarchical feature extraction), adaptive mu/sigma schedule (dynamic regularization per PnP iteration), SIREN INR coordinate output head (continuous representation for high-acceleration interpolation), joint LPIPS+SSIM denoiser training (perceptual+structural loss), dynamic PnP regularization scheduling (learnable lambda per iteration)
11	MMR-Mamba MMR-Mamba Zhao et al., Med. Image Anal. 2025 40.98 dB SSIM 0.969 Checkpoint unavailable	0.917	40.98	0.969	✓ Certified	Zhao et al., Med. Image Anal. 2025
12	BrainID-MRI BrainID-MRI Liu et al., CVPR 2025 41.0 dB SSIM 0.942 Checkpoint unavailable	0.904	41.0	0.942	✓ Certified	Liu et al., CVPR 2025
13	MRDynamo MRDynamo Chen et al., NeurIPS 2024 40.5 dB SSIM 0.938 Checkpoint unavailable	0.894	40.5	0.938	✓ Certified	Chen et al., NeurIPS 2024
14	MRI-DiffusionNet MRI-DiffusionNet Song et al., ICCV 2024 40.1 dB SSIM 0.932 Checkpoint unavailable	0.884	40.1	0.932	✓ Certified	Song et al., ICCV 2024
15	PromptMR PromptMR Bai et al., ECCV 2024 39.7 dB SSIM 0.926 Checkpoint unavailable	0.875	39.7	0.926	✓ Certified	Bai et al., ECCV 2024
16	E2E-VarNet E2E-VarNet Sriram et al., MICCAI 2020 39.4 dB SSIM 0.924 Checkpoint unavailable	0.869	39.4	0.924	✓ Certified	Sriram et al., MICCAI 2020
17	ReconFormer ReconFormer Guo et al., IEEE TMI 2024 39.0 dB SSIM 0.922 Checkpoint unavailable	0.861	39.0	0.922	✓ Certified	Guo et al., IEEE TMI 2024
18	HUMUS-Net HUMUS-Net Fabian et al., NeurIPS 2022 38.9 dB SSIM 0.923 Checkpoint unavailable	0.860	38.9	0.923	✓ Certified	Fabian et al., NeurIPS 2022
19	SwinMR SwinMR Huang et al., MICCAI 2022 38.5 dB SSIM 0.921 Checkpoint unavailable	0.852	38.5	0.921	✓ Certified	Huang et al., MICCAI 2022
20	HybridCascade HybridCascade Fastmri, arXiv 2020 37.8 dB SSIM 0.917 Checkpoint unavailable	0.839	37.8	0.917	✓ Certified	Fastmri, arXiv 2020
21	MoDL MoDL Aggarwal et al., IEEE TMI 2019 36.5 dB SSIM 0.912 Checkpoint unavailable	0.814	36.5	0.912	✓ Certified	Aggarwal et al., IEEE TMI 2019
22	U-Net U-Net Zbontar et al., arXiv 2018 35.9 dB SSIM 0.904 Checkpoint unavailable	0.800	35.9	0.904	✓ Certified	Zbontar et al., arXiv 2018
23	DCCNN DCCNN Schlemper et al., IEEE TMI 2018 35.5 dB SSIM 0.908 Checkpoint unavailable	0.796	35.5	0.908	✓ Certified	Schlemper et al., IEEE TMI 2018
24	Deep-ADMM-Net Deep-ADMM-Net Yang et al., NeurIPS 2016 35.3 dB SSIM 0.907 Checkpoint unavailable	0.792	35.3	0.907	✓ Certified	Yang et al., NeurIPS 2016
25	PnP-DnCNN PnP-DnCNN Ahmad et al., IEEE TCI 2019 (DOI:10.1109/TCI.2019.2944521) 35.0 dB SSIM 0.905 Try in SpecLab →	0.786	35.0	0.905	✓ Certified	Ahmad et al., IEEE TCI 2019 (DOI:10.1109/TCI.2019.2944521)
26	ALOHA ALOHA Jin et al., IEEE TMI 2016 34.5 dB SSIM 0.900 Try in SpecLab →	0.775	34.5	0.900	✓ Certified	Jin et al., IEEE TMI 2016
27	BM3D-MRI BM3D-MRI Eksioglu, Comput. Math. Meth. Med. 2016 34.2 dB SSIM 0.897 Try in SpecLab →	0.769	34.2	0.897	✓ Certified	Eksioglu, Comput. Math. Meth. Med. 2016
28	LORAKS LORAKS Haldar, IEEE TMI 2014 33.8 dB SSIM 0.893 Try in SpecLab →	0.760	33.8	0.893	✓ Certified	Haldar, IEEE TMI 2014
29	ESPIRiT ESPIRiT Uecker et al., MRM 2014 33.4 dB SSIM 0.890 Try in SpecLab →	0.752	33.4	0.890	✓ Certified	Uecker et al., MRM 2014
30	k-t SPARSE-SENSE k-t SPARSE-SENSE Lustig et al., MRM 2006 32.5 dB SSIM 0.875 Try in SpecLab →	0.729	32.5	0.875	✓ Certified	Lustig et al., MRM 2006
31	L1-Wavelet L1-Wavelet Lustig et al., MRM 2007 32.1 dB SSIM 0.870 Try in SpecLab →	0.720	32.1	0.870	✓ Certified	Lustig et al., MRM 2007
32	Score-MRI Score-MRI Chung & Ye, Med. Image Anal. 2022 31.4 dB SSIM 0.890 Checkpoint unavailable	0.718	31.4	0.890	✓ Certified	Chung & Ye, Med. Image Anal. 2022
33	GRAPPA GRAPPA Griswold et al., MRM 2002 31.2 dB SSIM 0.860 Try in SpecLab →	0.700	31.2	0.860	✓ Certified	Griswold et al., MRM 2002
34	SENSE SENSE Pruessmann et al., MRM 1999 29.5 dB SSIM 0.830 Try in SpecLab →	0.657	29.5	0.830	✓ Certified	Pruessmann et al., MRM 1999
35	Zero-Filled IFFT Zero-Filled IFFT Pruessmann et al., MRM 1999 26.0 dB SSIM 0.620 Try in SpecLab →	0.493	26.0	0.620	✓ Certified	Pruessmann et al., MRM 1999

Dataset: PWM Benchmark (35 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
🥇	SwinMR++ + gradient SwinMR++ + gradient Huang et al., IEEE TMI 2025 — multi-scale axial attention + INR head + k-space DC per module + LPIPS+SSIM+k-space joint loss + dynamic feature fusion Score 0.847 Correct & Reconstruct →	0.847	0.889 42.02 dB / 0.992	0.851 38.9 dB / 0.985	0.801 34.49 dB / 0.964	✓ Certified	Huang et al., IEEE TMI 2025 — multi-scale axial attention + INR head + k-space DC per module + LPIPS+SSIM+k-space joint loss + dynamic feature fusion
🥈	HUMUS-Net++ + gradient HUMUS-Net++ + gradient Fabian et al., dHUMUS-Net 2023 — k-space DC per module + dynamic multi-scale weighting + INR head + perceptual-structural loss + axial attention Score 0.838 Correct & Reconstruct →	0.838	0.901 41.63 dB / 0.991	0.823 37.15 dB / 0.979	0.789 33.6 dB / 0.958	✓ Certified	Fabian et al., dHUMUS-Net 2023 — k-space DC per module + dynamic multi-scale weighting + INR head + perceptual-structural loss + axial attention
🥉	MRI-FM + gradient MRI-FM + gradient Wang et al., Nature MI 2026 Score 0.824 Correct & Reconstruct →	0.824	0.889 40.51 dB / 0.989	0.809 35.17 dB / 0.969	0.773 32.13 dB / 0.944	✓ Certified	Wang et al., Nature MI 2026
4	PnP-DnCNN-Pro + gradient PnP-DnCNN-Pro + gradient PnP-DnCNN-Pro IEEE TMI 2025 (DOI:10.1109/TMI.2025.3441240) — multi-scale DnCNN denoiser + adaptive mu/sigma schedule + SIREN INR output head + joint LPIPS+SSIM denoiser training + dynamic PnP regularization scheduling Score 0.821 Correct & Reconstruct →	0.821	0.858 38.63 dB / 0.984	0.808 34.72 dB / 0.966	0.797 34.55 dB / 0.965	✓ Certified	PnP-DnCNN-Pro IEEE TMI 2025 (DOI:10.1109/TMI.2025.3441240) — multi-scale DnCNN denoiser + adaptive mu/sigma schedule + SIREN INR output head + joint LPIPS+SSIM denoiser training + dynamic PnP regularization scheduling
5	HybridCascade++ + gradient HybridCascade++ + gradient HybridCascade++ MICCAI 2021 + IEEE TMI 2025 — multi-scale cascade DC + SIREN INR warm-start + SSIM structural anchor + DRUNet polish + freq-blend LF/HF fusion Score 0.821 Correct & Reconstruct →	0.821	0.894 40.96 dB / 0.990	0.801 35.32 dB / 0.970	0.767 33.16 dB / 0.954	✓ Certified	HybridCascade++ MICCAI 2021 + IEEE TMI 2025 — multi-scale cascade DC + SIREN INR warm-start + SSIM structural anchor + DRUNet polish + freq-blend LF/HF fusion
6	MR-IPT + gradient MR-IPT + gradient Sci. Reports 2025 Score 0.815 Correct & Reconstruct →	0.815	0.873 39.62 dB / 0.987	0.803 35.34 dB / 0.970	0.769 33.61 dB / 0.958	✓ Certified	Sci. Reports 2025
7	HUMUS-Net + gradient HUMUS-Net + gradient Fabian et al., NeurIPS 2022 Score 0.812 Correct & Reconstruct →	0.812	0.832 36.05 dB / 0.974	0.815 34.82 dB / 0.967	0.789 34.18 dB / 0.962	✓ Certified	Fabian et al., NeurIPS 2022
8	U-Net++ + gradient U-Net++ + gradient Chen & Boning, IEEE TMI 2024 (DOI: 10.1109/TMI.2024.3367890) — Residual U-Net + data consistency layers + plug-and-play prior + residual connections + dense skip paths Score 0.811 Correct & Reconstruct →	0.811	0.862 38.76 dB / 0.985	0.800 34.79 dB / 0.966	0.771 31.84 dB / 0.941	✓ Certified	Chen & Boning, IEEE TMI 2024 (DOI: 10.1109/TMI.2024.3367890) — Residual U-Net + data consistency layers + plug-and-play prior + residual connections + dense skip paths
9	ReconFormer++ + gradient ReconFormer++ + gradient Pan et al., IEEE TMI 2025 Score 0.809 Correct & Reconstruct →	0.809	0.862 38.76 dB / 0.985	0.802 34.15 dB / 0.962	0.763 32.3 dB / 0.946	✓ Certified	Pan et al., IEEE TMI 2025
10	MMR-Mamba + gradient MMR-Mamba + gradient Zhao et al., Med. Image Anal. 2025 Score 0.802 Correct & Reconstruct →	0.802	0.879 39.97 dB / 0.988	0.784 33.24 dB / 0.955	0.744 31.09 dB / 0.932	✓ Certified	Zhao et al., Med. Image Anal. 2025
11	SwinMR + gradient SwinMR + gradient Huang et al., MICCAI 2022 Score 0.795 Correct & Reconstruct →	0.795	0.830 36.53 dB / 0.976	0.794 33.64 dB / 0.958	0.761 32.72 dB / 0.950	✓ Certified	Huang et al., MICCAI 2022
12	PromptMR-SFM + gradient PromptMR-SFM + gradient PWM 2026 Score 0.792 Correct & Reconstruct →	0.792	0.882 40.05 dB / 0.988	0.768 32.49 dB / 0.948	0.725 30.29 dB / 0.921	✓ Certified	PWM 2026
13	E2E-VarNet + gradient E2E-VarNet + gradient Sriram et al., MICCAI 2020 Score 0.787 Correct & Reconstruct →	0.787	0.838 36.56 dB / 0.976	0.774 32.92 dB / 0.952	0.750 31.37 dB / 0.935	✓ Certified	Sriram et al., MICCAI 2020
14	MoDL-Net++ + gradient MoDL-Net++ + gradient MoDL-Net++ IEEE TMI 2025 — multi-scale pyramid fusion + RDN/Swin deep prior + differentiable DC layers + LPIPS+SSIM+L1 joint loss + two-stage training strategy Score 0.785 Correct & Reconstruct →	0.785	0.866 39.72 dB / 0.987	0.757 32.25 dB / 0.945	0.733 30.39 dB / 0.923	✓ Certified	MoDL-Net++ IEEE TMI 2025 — multi-scale pyramid fusion + RDN/Swin deep prior + differentiable DC layers + LPIPS+SSIM+L1 joint loss + two-stage training strategy
15	ReconFormer + gradient ReconFormer + gradient Guo et al., IEEE TMI 2024 Score 0.784 Correct & Reconstruct →	0.784	0.856 37.92 dB / 0.982	0.792 33.95 dB / 0.960	0.704 28.18 dB / 0.885	✓ Certified	Guo et al., IEEE TMI 2024
16	MRDynamo + gradient MRDynamo + gradient Chen et al., NeurIPS 2024 Score 0.784 Correct & Reconstruct →	0.784	0.871 39.08 dB / 0.985	0.771 32.46 dB / 0.947	0.711 28.23 dB / 0.886	✓ Certified	Chen et al., NeurIPS 2024
17	PnP-DnCNN + gradient PnP-DnCNN + gradient Ahmad et al., IEEE TCI 2019 (DOI:10.1109/TCI.2019.2944521) Score 0.766 Correct & Reconstruct →	0.766	0.806 33.54 dB / 0.957	0.749 31.43 dB / 0.936	0.743 31.06 dB / 0.932	✓ Certified	Ahmad et al., IEEE TCI 2019 (DOI:10.1109/TCI.2019.2944521)
18	MRI-DiffusionNet + gradient MRI-DiffusionNet + gradient Song et al., ICCV 2024 Score 0.756 Correct & Reconstruct →	0.756	0.848 38.12 dB / 0.982	0.744 29.95 dB / 0.916	0.675 26.66 dB / 0.850	✓ Certified	Song et al., ICCV 2024
19	BrainID-MRI + gradient BrainID-MRI + gradient Liu et al., CVPR 2025 Score 0.755 Correct & Reconstruct →	0.755	0.856 38.29 dB / 0.983	0.737 30.81 dB / 0.928	0.671 27.39 dB / 0.867	✓ Certified	Liu et al., CVPR 2025
20	PromptMR + gradient PromptMR + gradient Bai et al., ECCV 2024 Score 0.753 Correct & Reconstruct →	0.753	0.844 37.91 dB / 0.982	0.746 30.88 dB / 0.929	0.668 26.27 dB / 0.839	✓ Certified	Bai et al., ECCV 2024
21	U-Net + gradient U-Net + gradient Zbontar et al., arXiv 2018 Score 0.746 Correct & Reconstruct →	0.746	0.796 33.36 dB / 0.956	0.754 31.08 dB / 0.932	0.687 28.27 dB / 0.886	✓ Certified	Zbontar et al., arXiv 2018
22	MoDL + gradient MoDL + gradient Aggarwal et al., IEEE TMI 2019 Score 0.733 Correct & Reconstruct →	0.733	0.804 34.3 dB / 0.963	0.723 28.71 dB / 0.895	0.671 26.29 dB / 0.840	✓ Certified	Aggarwal et al., IEEE TMI 2019
23	DCCNN + gradient DCCNN + gradient Schlemper et al., IEEE TMI 2018 Score 0.727 Correct & Reconstruct →	0.727	0.812 33.84 dB / 0.960	0.719 28.94 dB / 0.899	0.650 25.22 dB / 0.809	✓ Certified	Schlemper et al., IEEE TMI 2018
24	Deep-ADMM-Net + gradient Deep-ADMM-Net + gradient Yang et al., NeurIPS 2016 Score 0.717 Correct & Reconstruct →	0.717	0.809 33.76 dB / 0.959	0.703 28.16 dB / 0.884	0.640 24.71 dB / 0.793	✓ Certified	Yang et al., NeurIPS 2016
25	HybridCascade + gradient HybridCascade + gradient Fastmri, arXiv 2020 Score 0.714 Correct & Reconstruct →	0.714	0.818 34.95 dB / 0.967	0.698 27.68 dB / 0.874	0.626 25.3 dB / 0.812	✓ Certified	Fastmri, arXiv 2020
26	GRAPPA + gradient GRAPPA + gradient Griswold et al., MRM 2002 Score 0.703 Correct & Reconstruct →	0.703	0.726 28.86 dB / 0.898	0.713 28.8 dB / 0.897	0.671 27.08 dB / 0.860	✓ Certified	Griswold et al., MRM 2002
27	BM3D-MRI + gradient BM3D-MRI + gradient Eksioglu, Comput. Math. Meth. Med. 2016 Score 0.699 Correct & Reconstruct →	0.699	0.772 31.76 dB / 0.940	0.672 26.34 dB / 0.841	0.652 25.22 dB / 0.809	✓ Certified	Eksioglu, Comput. Math. Meth. Med. 2016
28	ESPIRiT + gradient ESPIRiT + gradient Uecker et al., MRM 2014 Score 0.696 Correct & Reconstruct →	0.696	0.786 32.31 dB / 0.946	0.657 25.44 dB / 0.816	0.646 25.51 dB / 0.818	✓ Certified	Uecker et al., MRM 2014
29	ALOHA + gradient ALOHA + gradient Jin et al., IEEE TMI 2016 Score 0.674 Correct & Reconstruct →	0.674	0.801 33.46 dB / 0.957	0.656 25.85 dB / 0.828	0.564 22.45 dB / 0.709	✓ Certified	Jin et al., IEEE TMI 2016
30	SENSE + gradient SENSE + gradient Pruessmann et al., MRM 1999 Score 0.672 Correct & Reconstruct →	0.672	0.692 26.93 dB / 0.856	0.681 26.58 dB / 0.848	0.644 26.15 dB / 0.836	✓ Certified	Pruessmann et al., MRM 1999
31	LORAKS + gradient LORAKS + gradient Haldar, IEEE TMI 2014 Score 0.660 Correct & Reconstruct →	0.660	0.790 32.61 dB / 0.949	0.612 23.69 dB / 0.757	0.578 22.92 dB / 0.728	✓ Certified	Haldar, IEEE TMI 2014
32	Score-MRI + gradient Score-MRI + gradient Chung & Ye, Med. Image Anal. 2022 Score 0.646 Correct & Reconstruct →	0.646	0.730 29.31 dB / 0.906	0.615 24.41 dB / 0.783	0.592 23.51 dB / 0.751	✓ Certified	Chung & Ye, Med. Image Anal. 2022
33	L1-Wavelet + gradient L1-Wavelet + gradient Lustig et al., MRM 2007 Score 0.613 Correct & Reconstruct →	0.613	0.743 30.3 dB / 0.921	0.586 23.14 dB / 0.737	0.511 19.91 dB / 0.594	✓ Certified	Lustig et al., MRM 2007
34	Zero-Filled IFFT + gradient Zero-Filled IFFT + gradient Pruessmann et al., MRM 1999 Score 0.604 Correct & Reconstruct →	0.604	0.644 24.36 dB / 0.781	0.574 22.29 dB / 0.702	0.594 23.3 dB / 0.743	✓ Certified	Pruessmann et al., MRM 1999
35	k-t SPARSE-SENSE + gradient k-t SPARSE-SENSE + gradient Lustig et al., MRM 2006 Score 0.604 Correct & Reconstruct →	0.604	0.742 29.59 dB / 0.910	0.582 22.31 dB / 0.703	0.489 19.9 dB / 0.594	✓ Certified	Lustig et al., MRM 2006

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	HUMUS-Net++ + gradient	0.901	41.63	0.991
2	HybridCascade++ + gradient	0.894	40.96	0.99
3	SwinMR++ + gradient	0.889	42.02	0.992
4	MRI-FM + gradient	0.889	40.51	0.989
5	PromptMR-SFM + gradient	0.882	40.05	0.988
6	MMR-Mamba + gradient	0.879	39.97	0.988
7	MR-IPT + gradient	0.873	39.62	0.987
8	MRDynamo + gradient	0.871	39.08	0.985
9	MoDL-Net++ + gradient	0.866	39.72	0.987
10	U-Net++ + gradient	0.862	38.76	0.985
11	ReconFormer++ + gradient	0.862	38.76	0.985
12	PnP-DnCNN-Pro + gradient	0.858	38.63	0.984
13	ReconFormer + gradient	0.856	37.92	0.982
14	BrainID-MRI + gradient	0.856	38.29	0.983
15	MRI-DiffusionNet + gradient	0.848	38.12	0.982
16	PromptMR + gradient	0.844	37.91	0.982
17	E2E-VarNet + gradient	0.838	36.56	0.976
18	HUMUS-Net + gradient	0.832	36.05	0.974
19	SwinMR + gradient	0.830	36.53	0.976
20	HybridCascade + gradient	0.818	34.95	0.967
21	DCCNN + gradient	0.812	33.84	0.96
22	Deep-ADMM-Net + gradient	0.809	33.76	0.959
23	PnP-DnCNN + gradient	0.806	33.54	0.957
24	MoDL + gradient	0.804	34.3	0.963
25	ALOHA + gradient	0.801	33.46	0.957
26	U-Net + gradient	0.796	33.36	0.956
27	LORAKS + gradient	0.790	32.61	0.949
28	ESPIRiT + gradient	0.786	32.31	0.946
29	BM3D-MRI + gradient	0.772	31.76	0.94
30	L1-Wavelet + gradient	0.743	30.3	0.921
31	k-t SPARSE-SENSE + gradient	0.742	29.59	0.91
32	Score-MRI + gradient	0.730	29.31	0.906
33	GRAPPA + gradient	0.726	28.86	0.898
34	SENSE + gradient	0.692	26.93	0.856
35	Zero-Filled IFFT + gradient	0.644	24.36	0.781

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
B0_inhomog	-1.5	3.0	ppm
gradient_nonlin	-2.0	4.0	%
coil_sensitivity	-5.0	10.0	%
k_trajectory	-1.0	2.0	%

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	SwinMR++ + gradient	0.851	38.9	0.985
2	HUMUS-Net++ + gradient	0.823	37.15	0.979
3	HUMUS-Net + gradient	0.815	34.82	0.967
4	MRI-FM + gradient	0.809	35.17	0.969
5	PnP-DnCNN-Pro + gradient	0.808	34.72	0.966
6	MR-IPT + gradient	0.803	35.34	0.97
7	ReconFormer++ + gradient	0.802	34.15	0.962
8	HybridCascade++ + gradient	0.801	35.32	0.97
9	U-Net++ + gradient	0.800	34.79	0.966
10	SwinMR + gradient	0.794	33.64	0.958
11	ReconFormer + gradient	0.792	33.95	0.96
12	MMR-Mamba + gradient	0.784	33.24	0.955
13	E2E-VarNet + gradient	0.774	32.92	0.952
14	MRDynamo + gradient	0.771	32.46	0.947
15	PromptMR-SFM + gradient	0.768	32.49	0.948
16	MoDL-Net++ + gradient	0.757	32.25	0.945
17	U-Net + gradient	0.754	31.08	0.932
18	PnP-DnCNN + gradient	0.749	31.43	0.936
19	PromptMR + gradient	0.746	30.88	0.929
20	MRI-DiffusionNet + gradient	0.744	29.95	0.916
21	BrainID-MRI + gradient	0.737	30.81	0.928
22	MoDL + gradient	0.723	28.71	0.895
23	DCCNN + gradient	0.719	28.94	0.899
24	GRAPPA + gradient	0.713	28.8	0.897
25	Deep-ADMM-Net + gradient	0.703	28.16	0.884
26	HybridCascade + gradient	0.698	27.68	0.874
27	SENSE + gradient	0.681	26.58	0.848
28	BM3D-MRI + gradient	0.672	26.34	0.841
29	ESPIRiT + gradient	0.657	25.44	0.816
30	ALOHA + gradient	0.656	25.85	0.828
31	Score-MRI + gradient	0.615	24.41	0.783
32	LORAKS + gradient	0.612	23.69	0.757
33	L1-Wavelet + gradient	0.586	23.14	0.737
34	k-t SPARSE-SENSE + gradient	0.582	22.31	0.703
35	Zero-Filled IFFT + gradient	0.574	22.29	0.702

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
B0_inhomog	-1.8	2.7	ppm
gradient_nonlin	-2.4	3.6	%
coil_sensitivity	-6.0	9.0	%
k_trajectory	-1.2	1.8	%

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	SwinMR++ + gradient	0.801	34.49	0.964
2	PnP-DnCNN-Pro + gradient	0.797	34.55	0.965
3	HUMUS-Net++ + gradient	0.789	33.6	0.958
4	HUMUS-Net + gradient	0.789	34.18	0.962
5	MRI-FM + gradient	0.773	32.13	0.944
6	U-Net++ + gradient	0.771	31.84	0.941
7	MR-IPT + gradient	0.769	33.61	0.958
8	HybridCascade++ + gradient	0.767	33.16	0.954
9	ReconFormer++ + gradient	0.763	32.3	0.946
10	SwinMR + gradient	0.761	32.72	0.95
11	E2E-VarNet + gradient	0.750	31.37	0.935
12	MMR-Mamba + gradient	0.744	31.09	0.932
13	PnP-DnCNN + gradient	0.743	31.06	0.932
14	MoDL-Net++ + gradient	0.733	30.39	0.923
15	PromptMR-SFM + gradient	0.725	30.29	0.921
16	MRDynamo + gradient	0.711	28.23	0.886
17	ReconFormer + gradient	0.704	28.18	0.885
18	U-Net + gradient	0.687	28.27	0.886
19	MRI-DiffusionNet + gradient	0.675	26.66	0.85
20	BrainID-MRI + gradient	0.671	27.39	0.867
21	MoDL + gradient	0.671	26.29	0.84
22	GRAPPA + gradient	0.671	27.08	0.86
23	PromptMR + gradient	0.668	26.27	0.839
24	BM3D-MRI + gradient	0.652	25.22	0.809
25	DCCNN + gradient	0.650	25.22	0.809
26	ESPIRiT + gradient	0.646	25.51	0.818
27	SENSE + gradient	0.644	26.15	0.836
28	Deep-ADMM-Net + gradient	0.640	24.71	0.793
29	HybridCascade + gradient	0.626	25.3	0.812
30	Zero-Filled IFFT + gradient	0.594	23.3	0.743
31	Score-MRI + gradient	0.592	23.51	0.751
32	LORAKS + gradient	0.578	22.92	0.728
33	ALOHA + gradient	0.564	22.45	0.709
34	L1-Wavelet + gradient	0.511	19.91	0.594
35	k-t SPARSE-SENSE + gradient	0.489	19.9	0.594

Spec Ranges (4 parameters)

Parameter	Min	Max	Unit
B0_inhomog	-1.05	3.45	ppm
gradient_nonlin	-1.4	4.6	%
coil_sensitivity	-3.5	11.5	%
k_trajectory	-0.7	2.3	%

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

MRI forms images by exciting hydrogen nuclei with RF pulses in a strong magnetic field (1.5-7T) and measuring the emitted RF signal with receive coils. Spatial encoding uses gradient fields to map signal frequency and phase to spatial position, acquiring data in k-space (spatial frequency domain). The forward model for parallel imaging is y_c = F_u * S_c * x + n_c where F_u is the undersampled Fourier transform, S_c are coil sensitivity maps, and n_c is complex Gaussian noise. Accelerated MRI undersamples k-space (4-8x) and uses SENSE, GRAPPA, or deep-learning (E2E-VarNet) for reconstruction.

Principle

Magnetic Resonance Imaging measures the precession of hydrogen nuclear spins in a strong magnetic field (1.5-7 T). Radiofrequency pulses tip spins away from equilibrium, and gradient fields spatially encode the MR signal into k-space (spatial frequency domain). The image is obtained by inverse Fourier transform of k-space data. Contrast depends on tissue T1, T2, and proton density via the pulse sequence timing parameters.

How to Build the System

A clinical MRI scanner has a superconducting magnet (1.5 T or 3 T), gradient coils (40-80 mT/m, 200 T/m/s slew rate), RF transmit body coil, and local receive coil arrays (8-128 channels). The patient lies inside the bore on a table. Key calibrations: center frequency, RF transmit calibration (B₁ mapping), shimming (B₀ homogeneity), and gradient eddy current compensation. Use pulse sequences optimized for the clinical question (T1w, T2w, FLAIR, DWI, etc.).

Common Reconstruction Algorithms

Inverse FFT (standard Cartesian k-space reconstruction)
GRAPPA (GeneRalized Autocalibrating Partially Parallel Acquisitions)
SENSE (SENSitivity Encoding) parallel imaging
Compressed sensing MRI (L1-wavelet + TV regularization)
Deep-learning MRI reconstruction (fastMRI, variational networks, E2E-VarNet)

Common Mistakes

Aliasing artifacts from insufficient FOV or acceleration too aggressive
Motion artifacts (ghosting in phase-encode direction) from patient or physiological motion
B₀ inhomogeneity causing geometric distortion and signal dropout (especially at 3T+)
Fat-water chemical shift artifacts at fat-tissue interfaces
Incorrect coil sensitivity maps causing SENSE/GRAPPA reconstruction artifacts

How to Avoid Mistakes

Set FOV to cover the anatomy with margin; use saturation bands to suppress aliasing
Apply motion correction (navigator, PROPELLER, prospective correction) for moving anatomy
Perform careful shimming; use distortion correction maps for EPI sequences
Use fat suppression or water-fat separation (Dixon) sequences
Acquire adequate auto-calibration data for parallel imaging; use robust coil maps

Forward-Model Mismatch Cases

The widefield fallback produces real-valued spatially blurred output, but MRI acquires complex-valued k-space data via the Fourier transform with undersampling mask — all phase information is lost with the fallback
The fallback applies spatial-domain convolution, but MRI measurement occurs in Fourier domain (k-space): y = M * F * x — using the fallback means compressed-sensing MRI reconstruction (L1-wavelet, E2E-VarNet) cannot function

How to Correct the Mismatch

Use the MRI operator that applies the 2D Fourier transform followed by an undersampling mask: y = M * FFT2(x), producing complex-valued k-space measurements
Reconstruct using parallel imaging (GRAPPA, SENSE) or compressed sensing (L1-wavelet + TV regularization) that operate on the Fourier-domain measurements with known sampling pattern

Experimental Setup — Signal Chain

Experimental setup diagram for Magnetic Resonance 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

16.369188751529634 dB SSIM 0.3769596313670564

II — Mismatch

Ground Truth

Measurement

Reconstruction

28.91599361455386 dB SSIM 0.8937325593738555

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

43.0888232643681 dB SSIM 0.9970601105022431

I — Ideal

Ground Truth

Measurement

Reconstruction

17.02984109629589 dB SSIM 0.5076141905187631

II — Mismatch

Ground Truth

Measurement

Reconstruction

29.54697836453365 dB SSIM 0.9088858192167282

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

44.00498178524224 dB SSIM 0.9974331602373123

I — Ideal

Ground Truth

Measurement

Reconstruction

19.876634796830935 dB SSIM 0.5132084125883123

II — Mismatch

Ground Truth

Measurement

Reconstruction

29.296469693787934 dB SSIM 0.8901788054246903

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

45.80916505985154 dB SSIM 0.9983439806976319

I — Ideal

Ground Truth

Measurement

Reconstruction

16.209230008064964 dB SSIM 0.3451008090497886

II — Mismatch

Ground Truth

Measurement

Reconstruction

28.211197909256423 dB SSIM 0.8836538261470794

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

42.81516446798818 dB SSIM 0.997371546825409

Mean PSNR Across All Scenes

17.371223663180356 dB

I — Ideal

28.99265989553297 dB

II — Mismatch

43.929533644362515 dB

III — Oracle

Degradation: 11.6 dB Recovery: +14.9 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	16.369188751529634	0.3769596313670564	28.91599361455386	0.8937325593738555	43.0888232643681	0.9970601105022431
scene_01	17.02984109629589	0.5076141905187631	29.54697836453365	0.9088858192167282	44.00498178524224	0.9974331602373123
scene_02	19.876634796830935	0.5132084125883123	29.296469693787934	0.8901788054246903	45.80916505985154	0.9983439806976319
scene_03	16.209230008064964	0.3451008090497886	28.211197909256423	0.8836538261470794	42.81516446798818	0.997371546825409
Mean	17.371223663180356	0.4357207608809801	28.99265989553297	0.8941127525405883	43.929533644362515	0.9975521995656491

Experimental Setup

Instrument: Siemens MAGNETOM Prisma / GE SIGNA Premier 3T

Anatomy: knee / brain

Matrix Size: 320x320

Field Strength T: 3.0

Receive Coils: 15

Acceleration Factor: 4

K Space Sampling: variable-density random Cartesian

Center Fraction: 0.08

Sequence: TSE (turbo spin echo)

Reconstruction: SENSE / E2E-VarNet

Dataset: fastMRI (knee: 1594, brain: 6970 volumes)

Key References

Pruessmann et al., 'SENSE: Sensitivity encoding for fast MRI', Magnetic Resonance in Medicine 42, 952-962 (1999)
Zbontar et al., 'fastMRI: An open dataset and benchmarks for accelerated MRI', arXiv:1811.08839 (2018)
Sriram et al., 'End-to-End Variational Networks for Accelerated MRI Reconstruction (E2E-VarNet)', MICCAI 2020

Canonical Datasets

fastMRI (knee: 1594 volumes, brain: 6970 volumes)
Calgary-Campinas (brain, multi-coil)
SKM-TEA (Stanford knee MRI)

Spec DAG — Forward Model Pipeline

F(k-traj) → D(g, η₁)

F k-Space Sampling (k-traj)

D RF Coil Receiver (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
ΔB₀	B0_inhomog	B₀ field inhomogeneity (ppm)	1.5
ΔG	gradient_nonlin	Gradient nonlinearity (%)	2.0
ΔS	coil_sensitivity	Coil sensitivity map error (%)	5.0
Δk	k_trajectory	k-space trajectory error (%)	1.0

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.