Inaccuracies and temporal fluctuations in field map measurements form a major problem in image reconstruction for permanent magnet based low field MRI systems. These inaccuracies can potentially be corrected by using a joint image reconstruction and field map estimation algorithm. Simulation results show improved image quality when using a new updating scheme compared to standard iterative reconstructions.
Experimental setup: A Halbach magnet was constructed, containing four rings each with 24 2.5×2.5×2.5 cm3 cubes of N-52 neodymium boron iron magnets (producing a 0.06T magnetic field at the center), with an internal ring of smaller magnets to produce a linear field gradient. The magnetic field was measured on a 5×7.5 mm2 grid using a gaussmeter (Alphalab GM-2).
Simulation: A solenoid coil was modelled for RF transmission and reception. The sample was rotated 36 times with angular increments of $$$\theta =10$$$°. For each rotation, a spin-echo sequence was simulated with RF pulse durations of 20 µs, and TE=5 ms; 512 data points and a dwell time of 2 µs. A signal model
$$\textbf{S}_\theta=ETCR_\theta \textbf{m} \quad \text{(1)}$$
was used with $$$\textbf{m}$$$ the unknown image (simulated as a Shepp Logan MATLAB phantom), $$$\textbf{S}_\theta$$$ the measured time-domain signal for rotation angle $$$\theta$$$, $$$R_\theta$$$ the corresponding rotation matrix, $$$C$$$ a diagonal matrix with uniform spatial coil sensitivity weighted by the coil frequency profile (Q-factor of 13.8 based on an S11 measurement) on the diagonal, $$$T$$$ a diagonal matrix including the pulse profile (simulated via the Bloch equations) and transmit bandwidth, $$$E$$$ the signal encoding matrix with elements $$$E_{pq}=e^{-2\pi iB_q t_p}$$$, and $$$B_q$$$ the main field difference (with respect to the transmitter frequency) in Hz in pixel $$$q$$$. White Gaussian noise was added to the signals such that the SNR was 10.
Image reconstruction: The image is reconstructed by solving the non-linear problem
$$\hat{\textbf{m}}=\min \frac{\mu}{2}|| \sum_{\theta} \textbf{S}_\theta - ETCR_\theta \textbf{m} ||_2^2 + \frac{\lambda}{2} TV(\textbf{m}) \quad \text{(2)}$$
where $$$TV$$$ is a total variation operator and $$$\mu$$$ and $$$\lambda$$$ regularization parameters. Substitution of the perturbations $$$\textbf{m}^R=\textbf{m}^R+\Delta \textbf{m}^R $$$, $$$\textbf{m}^I=\textbf{m}^I+\Delta \textbf{m}^I $$$ and $$$\textbf{B}=\textbf{B}+\Delta \textbf{B}$$$ in Eq. (1) gives a system for the error terms:
$$\sum_\theta A_\theta^H \left[ \begin{matrix} \textbf{S}_\theta^R -\textbf{a}_1 \\ \textbf{S}_\theta^I - \textbf{a}_2 \end{matrix} \right] = \sum_\theta A_\theta^H A_\theta \left[ \begin{matrix} \Delta \textbf{m}^R \\ \Delta \textbf{m}^I \\ \Delta \textbf{B} \end{matrix} \right] \quad \text{(3)} $$
$$$\textbf{S}_\theta^R/\textbf{S}_\theta^I$$$ are the real/imaginary parts of the sampled time signal at rotation angle $$$\theta$$$. $$$A_\theta$$$ describes the linearized data model for the error terms and $$$\textbf{a}_1/\textbf{a}_2$$$ describe the simulated time domain signal based on the estimated image $$$\textbf{m}$$$. The final image is reconstructed by the algorithm shown in Figure 1.
1. Cooley C, Stockmann J, Armstrong B, et al. 2D Imaging in a Lightweight Portable MRI Scanner without Gradient Coils. Magnetic Resonance in Medicine. 2015;72(3):1–15.
2. Ren Z, Obruchkov S, Lu D, et al. A Low-field Portable Magnetic Resonance Imaging System for Head Imaging. Progress in Electromagnetics Research Symposium. 2017:3042–3044.
3. Cooley C, Haskell M, Cauley S, et al. Design of Sparse Halbach Magnet Arrays for Portable MRI Using a Genetic Algorithm. IEEE Transactions on Magnetics. 2017;54(1):1–12.
4. Reeder S, Wen Z, Yu H, et al. Multicoil Dixon Chemical Species Separation With an Iterative Least-Squares Estimation Method. Magnetic Resonance in Medicine. 2004;51(1):35–45.