| Title:
|
Image Reconstruction methods for Low-Field MRI with Severe Field Inhomogeneity in a Halbach magnet
|
| Author(s):
|
1
|
| Subject:
|
Advanced Techniques in Signal Acquisition and Image Processing
|
|
|
|
Presenter Author:
|
|
| Name:
|
Fatemeh Alirezaee
|
|
|
|
Author:
|
|
| Name 1:
|
Fatemeh Alirezaee - Corresponding - Presenter
|
| Affiliation 1:
|
Medical Radiation Engineering Department - Shahid Beheshti University
|
|
|
| Name 2:
|
Mohammad Mohammadzadeh
|
| Affiliation 2:
|
Medical Radiation Engineering Department - Shahid Beheshti University
|
|
|
| Name 3:
|
|
| Affiliation 3:
|
|
|
|
| Name 4:
|
|
| Affiliation 4:
|
|
|
|
| Name 5:
|
|
| Affiliation 5:
|
|
|
|
| Name 6:
|
|
| Affiliation 6:
|
|
|
|
| Name 7:
|
|
| Affiliation 7:
|
|
|
|
| Name 8:
|
|
| Affiliation 8:
|
|
|
|
| Name 9:
|
|
| Affiliation 9:
|
|
|
|
| Name 10:
|
|
| Affiliation 10:
|
|
|
|
Abstract
|
Original Research
|
|
introduction :
|
Low-field MRI systems based on Halbach arrays offer advantages in cost, portability, and accessibility, but they suffer from strong field (B0) inhomogeneities even after optimization, particularly in magnets with low aspect ratios (length-to-inner-diameter). These inhomogeneities cause signal loss and produce spatially varying off-resonance frequencies in the kilohertz range, leading to geometric image distortions in standard Fourier-based reconstruction. Several reconstruction strategies have been proposed to mitigate B0-induced distortions. Conjugate phase reconstruction (CPR) compensates for off-resonance phase accumulation during readout and can partially reduce artifacts for moderate field variations (Noll, 1991). Model-based (MB) reconstruction enable high-fidelity reconstruction under severe inhomogeneity (Fessler, 2010; Liesker, 2021). More recently, physics-informed deep learning (DL) approaches have emerged as efficient alternatives that combine data-driven modeling with physical constraints (Bekkering, 2021; Schote, 2024; Chen, 2024). In this study, we systematically evaluate FFT, CPR, MB, and DL methods for distortion correction in a simulated low-field MRI system based on an optimized Halbach magnet for adult head imaging, with an aspect ratio of 1.1, 〈B0〉=54mT, and magnetic field inhomogeneity up to 10000ppm over a DSV:20cm. Using a three-dimensional magnetic field map, a time-shifted fast spin echo acquisition, and a Shepp–Logan phantom, reconstruction performance is quantified using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean squared error (MSE) metrics.
|
|
Methods:
|
A three-dimensional magnetic field map was provided with 4×4×4mm³ resolution. For two-dimensional imaging, an axial slice at x=0 was extracted, consistent with the imaging geometry. The resulting B0 map was converted to a spatially varying off-resonance frequency map Δf(r)=γΔB(r) and interpolated onto a 128×128 grid covering a 0.2m field of view (FOV). A Shepp–Logan phantom resized to the same grid served as the ground-truth object. Synthetic k-space data were generated using an off-resonance signal model and a time-shifted FSE acquisition. FFT was used as a baseline. CPR applied B0-derived phase correction, and MB solved the inverse problem iteratively using a full encoding model incorporating the B₀ map. The DL approach employed a physics-informed network trained on simulated data, enforcing consistency with the MRI signal model. Reconstruction quality was evaluated using PSNR, SSIM, and MSE.
|
|
Results:
|
FFT exhibited severe distortion, blurring, and signal pile-up. CPR reduced phase-related artifacts but failed to correct spatial misregistration in regions with high inhomogeneity. Both MB and DL improved image sharpness and intensity uniformity. Quantitative metrics ranked MB and DL best, followed by CPR and FFT.
|
|
Conclusion:
|
This study demonstrates that FFT is inadequate for systems with severe B₀ inhomogeneity. Under kilohertz-order off-resonance conditions, FFT fails to preserve geometric accuracy and signal fidelity. CPR provides modest improvement by compensating for global phase errors, but its effectiveness degrades when off-resonance effects approach or exceed the readout bandwidth. MB achieves near-optimal distortion correction by explicitly modeling the full encoding process. However, this accuracy costs high computational complexity and sensitivity to field-map accuracy, which may restrict its use in real-time or resource-constrained settings. Physics-informed DL achieves image quality comparable to MB while offering significantly reduced inference time after training, making it a compelling alternative for portable low-field MRI systems.
|
| Attachment:
|
|
| Keywords:
|
MRI Image Reconstruction, Low field MRI system, Halbach Array Magnet, magnetic field inhomogeneity, Deep learning image reconstruction, image distrotion correction
|
| Key message:
|
|
| Do you want your abstract to be indexed, if this is selected as one of the top 100
articles?
|
