Magnetic Resonance Imaging
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.
Fourier Undersampling
Gaussian
sense
RF_COIL
Forward-Model Signal Chain
Each primitive represents a physical operation in the measurement process. Arrows show signal flow left to right.
F(k-traj) → D(g, η₁)
Benchmark Variants & Leaderboards
MRI
Magnetic Resonance Imaging
F(k-traj) → D(g, η₁)
Standard Leaderboard (Top 10)
| # | Method | Score | PSNR (dB) | SSIM | Trust | Source |
|---|---|---|---|---|---|---|
| 🥇 | SwinMR++ | 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++ | 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 | 0.950 | 42.48 | 0.983 | ✓ Certified | Sci. Reports 2025 |
| 4 | HybridCascade++ | 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++ | 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++ | 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 | 0.926 | 42.1 | 0.948 | ✓ Certified | Wang et al., Nature MI 2026 |
| 8 | ReconFormer++ | 0.926 | 41.5 | 0.969 | ✓ Certified | Pan et al., IEEE TMI 2025 |
| 9 | PromptMR-SFM | 0.924 | 41.3 | 0.971 | ✓ Certified | PWM 2026 |
| 10 | PnP-DnCNN-Pro | 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) |
Showing top 10 of 35 methods. View all →
Mismatch Parameters (4) click to expand
| Name | Symbol | Description | Nominal | Perturbed |
|---|---|---|---|---|
| B0_inhomog | ΔB₀ | B₀ field inhomogeneity (ppm) | 0 | 1.5 |
| gradient_nonlin | ΔG | Gradient nonlinearity (%) | 0 | 2.0 |
| coil_sensitivity | ΔS | Coil sensitivity map error (%) | 0 | 5.0 |
| k_trajectory | Δk | k-space trajectory error (%) | 0 | 1.0 |
Reconstruction Triad Diagnostics
The three diagnostic gates (G1, G2, G3) characterize how reconstruction quality degrades under different error sources. Each bar shows the relative attribution.
Model: fourier undersampling — Mismatch modes: off resonance, motion, eddy current, coil sensitivity error, trajectory error
Noise: gaussian — Typical SNR: 15.0–35.0 dB
Requires: coil sensitivity maps, field inhomogeneity, k space trajectory, noise covariance
Modality Deep Dive
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
Siemens MAGNETOM Prisma / GE SIGNA Premier 3T
knee / brain
320x320
3.0
15
4
variable-density random Cartesian
0.08
TSE (turbo spin echo)
SENSE / E2E-VarNet
fastMRI (knee: 1594, brain: 6970 volumes)
Signal Chain Diagram
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)