Ultra-low-field (ULF) Magnetic Resonance Imaging (MRI) offers a portable, low-cost alternative to conventional scanners by operating at field strengths below 10 mT. At these fields, two sensor technologies compete: quantum SQUID (Superconducting Quantum Interference Device) sensors, which provide high sensitivity but require cryogenic cooling, and classical induction coils, which are inexpensive and room-temperature but produce weaker signals. With conventional physics-based imaging, both suffer from low signal-to-noise ratios (SNR). Data-driven reconstruction has overcome similar limits in related inverse problems, but its use for ULF-MRI has been hindered by a lack of open datasets, benchmarks, and flexible simulators. To close this gap, we present OpenULF, a modular, open-source simulation toolkit that generates paired training data and supports end-to-end machine-learning (ML) reconstruction. OpenULF integrates a Bloch-equation-validated physics simulator, calibrated SQUID and coil receiver models, reconstruction ML networks, and an automated design-space exploration engine. Architecturally, a receiver abstract base class, a typed HDF5 contract that decouples physics from learning, and a resumable sweep engine make sensors, models, and datasets independently substitutable. Using OpenULF, we train 774 models across four anatomical datasets and conduct three design-space explorations. The effectiveness of ML reconstruction depends strongly on the dataset and the sensor. We identify a SQUID-to-coil crossover near 450 μT and demonstrate that sensor hardware and ML model capacity can be traded off during system design. Sensor and reconstruction choices must therefore be co-designed: the best sensor is set by the dataset and field strength, not by SNR.
Figures & tables
Feature
MRiLab
JEMRIS
fastMRI
OpenULF
Bloch spin dynamics
✓
✓
–
✓
Pulse sequence sim.
✓
✓
–
✓
ULF sensor models
–
–
–
✓
SQUID + Coil in one
–
–
–
✓
ML reconstruction
–
–
✓
✓
Multi-model library
–
–
✓
✓
TABLE I : Feature comparison of MRI simulation tools (✓ = supported, – = not). “Automated param. sweep” means a built-in resumable, deduplicated sweep engine, not the general scripting that MRiLab and JEMRIS also provide; “Multi-model library” means bundled reconstruction models.
Fig. 1 : Overview of OpenULF. (a) Component library with extensible physics, ML, data, and composition modules. (b) OpenULF toolkit: the co-simulation pipeline (left sub-box) transforms tissue maps through sensor-specific receivers and ML reconstruction; the design-space exploration engine (right sub-box) sweeps B0×NEX×channels×model×sensor , training 774 models across four datasets. (c) Design insights: three quantitative design rules plus cross-dataset and compression-ratio validation. The top-level feedback arrow indicates that insights drive iterative extension of the component library.
#
Test description
Error
1
Single-voxel M⊥ vs. spin-echo (6 tissues)
0.00%
2
T1 contrast: GM/WM ratio vs. TR (5 pts)
0.00%
3
T2 decay: signal vs. TE (6 points)
0.00%
4
Prepolarization (12 config combinations)
< 0.01%
5
Image: Parseval, k-space DC, tissue order
0.00%
6
T1 field dependence [ 23 ]
0.00%
TABLE II : Numerical-correctness check of the forward engine against closed-form Bloch solutions. Near-zero error is expected of a correct implementation; physical realism comes from measured parameters (Section IV-A ). Here 0.00% denotes error below the two-decimal display threshold.
SQUID
Coil
Both
Dataset
Model
dB
SSIM
dB
SSIM
dB
SSIM
Gap
BrainWeb
Input
11.0
.16
5.3
.02
–
–
+5.8
SimpleCNN
18.0
.41
12.2
.16
18.1
.49
+5.9
DnCNN
18.9
.66
13.8
.37
19.0
.68
+5.2
UNet
19.8
.80
16.5
.59
19.8
.79
+3.4
NAFNet
19.8
.78
16.5
.61
19.8
.78
+3.3
TABLE III : Test-set PSNR (dB) and SSIM across 4 datasets, 5 models, and 3 sensor configurations. The Input row reports the raw sensor output (before ML) against the ideal image. Gap = SQUID − Coil in PSNR.
Fig. 2 : Reconstruction on a single representative BrainWeb test slice (per-slice PSNR shown; test-set means are in Table III ). Top row: ideal target and noisy inputs from SQUID and coil. Bottom row: UNet outputs from each sensor configuration.
Fig. 3 : Input vs. output SQUID-coil PSNR gap for four datasets (marker: 5-model mean, whisker: min–max across the five networks). The dashed CR = 1 line is where ML leaves the gap unchanged; above it the gap is amplified, below compressed.
Fig. 4 : PSNR gap (SQUID minus coil) vs. B0 for three models at NEX = 50. Positive values indicate SQUID is the superior sensor; negative values favor the coil. All three models cross zero near 450 μ T; the shaded band (418–473 μ T) spans the three models’ individual zero-crossings, confirming that the crossover is driven by sensor physics rather than model choice.
Fig. 5 : Design-space coverage: test PSNR (UNet) over the swept (B0,NEX) grid for the SQUID (left) and coil (right) sensors. Brighter is better; gray cells were not trained, exposing the sweep’s goal-directed, non-exhaustive coverage. SQUID leads at low B0 and the coil at high B0 ; the crossover of Fig. 4 is where they switch.
Fig. 6 : Two faces of hardware–ML substitutability on SQUID ( B0=50μ T). Left: test PSNR vs. NEX, where UNet at NEX = 4 matches SimpleCNN at NEX = 20, trading model capacity for a 5× shorter scan. Right: test PSNR vs. SQUID channel count, where a 2-channel UNet matches SimpleCNN between its 3- and 4-channel points, trading model capacity for sensor channels.
Ultra-low-field (ULF) MRI offers portable and accessible neuroimaging but suffers from reduced signal-to-noise ratio and limited spatial resolution compared to high-field (HF) systems. Acquiring paired ULF-HF data for supervised enhancement is often difficult, particularly in resource-limited settings. We introduce ULF-Synth, a framework that combines: (i) acquisition-based synthesis of realistic ULF images from HF volumes to create large-scale paired training data, (ii) a spatial-frequency domain objective that prioritizes recovery of high-frequency anatomical detail. This formulation is architecture-agnostic, consistently improving structural similarity and perceptual fidelity across encoder-decoder, adversarial, and diffusion-based translation models. When trained exclusively on synthetic data, the resulting models generalize effectively to real 64mT ULF acquisitions, improving downstream multiclass brain segmentation and achieving higher radiologist preference and diagnostic acceptability in a blinded reader study. These findings demonstrate that synthetic paired supervision provides a practical and scalable pathway for enhancing ULF MRI without requiring real paired acquisitions. Code, Models and Dataset: https://github.com/toufiqmusah/ULF-Synth
Kwame Nkrumah University of Science and Technology, Kumasi, Ghana · University of Catania, Catania, Italy · The Hong Kong University of Science and Technology, Hong Kong SAR, China +1
Ultra-low-field (ULF) MRI offers portable and low-cost imaging but suffers from poor image quality. To address this, we present our submission to the 2025 ULF Enhancement Challenge (ULF-EnC), where the goal is to synthesise high-field-like MRIs from 64 mT scans. Our pipeline enhances ULF MRI through a combination of anatomical conditioning and model ensembling. We first generate tissue segmentation priors using a Swin UNETR trained solely on challenge-provided data. These priors condition two independent enhancement networks - a CycleGAN and a transformer-based residual enhancement model (T-REX) - each trained to synthesise 3 T-like MRIs. Outputs from both models are combined using a weighted average. Our approach produces enhanced MRIs that were comparable to high-field scans both quantitatively and qualitatively.
James Grover, Andrew Phair, Michael Ferraro +1
Image X Institute, Sydney School of Health Sciences, Faculty of Medicine and Health, The University of Sydney
Low-field (LF) magnetic resonance imaging (MRI) improves accessibility and reduces costs but generally has lower signal-to-noise ratios and degraded contrast compared to high field (HF) MRI, limiting its clinical utility. Simulating LF MRI from HF MRI enables virtual evaluation of novel imaging devices and development of LF algorithms. Existing low field simulators rely on noise injection and smoothing, which fail to capture the contrast degradation seen in LF acquisitions. To this end, we introduce an end-to-end LF-MRI synthesis framework that learns HF to LF image degradation directly from a small number of paired HF-LF MRIs. Specifically, we introduce a novel HF to LF coordinate-image decoupled neural operator (H2LO) to model the underlying degradation process, and tailor it to capture high-frequency noise textures and image structure. Experimental results in T1w and T2w MRI demonstrate that H2LO produces more faithful simulated low-field images than existing parameterized noise synthesis models and popular image-to-image translation models. Furthermore, it improves performance in downstream image enhancement tasks, showcasing its potential to enhance LF MRI diagnostic capabilities.
Ziqi Gao, Nicha Dvornek, Xiaoran Zhang +3
Departments of Biomedical Engineering, Yale University · Department of Radiology & Biomedical Imaging, Yale University