Portable Magnet Design Optimization for Brain Imaging without Gradient Coils
Clarissa Zimmerman Cooley1, Melissa Haskell1,2, Jason P Stockmann1, Cristen Lapierre1, Chenoa Schatzki-McClain1, Charlotte Sappo1, Stephen F Cauley1,3, Bastien Guerin1,3, Matthew S Rosen1,3,4, and Lawrence L Wald1,3,5

1A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States, 2Biophysics, Harvard University, Cambridge, MA, United States, 3Harvard Medical School, Boston, MA, United States, 4Department of Physics, Harvard University, Cambridge, MA, United States, 5Harvard-MIT Division of Health Sciences Technology, Cambridge, MA, United States

Synopsis

The development of a low-cost portable MRI scanner for brain imaging could address the need for imaging at unconventional sites. We previously presented an appropriate method for 3D imaging in a portable magnet without gradient coils. In-plane image encoding was demonstrated using the natural field variation of a rotating prototype magnet. However, the magnet diameter must be increased for human brain imaging and the built-in encoding field should be tailored to improve image resolution uniformity. We present a method for designing an optimized magnet for this application using the Genetic Algorithm, and evaluate a chosen design via image simulations.

Purpose

To design a head-size permanent magnet with built-in Spatial Encoding Magnetic fields (SEM) to support a portable brain imaging system without gradient coils.

Methods

3D MR imaging has been shown in a portable rotating Halbach cylinder magnet1, where generalized projection imaging is performed in the y-z plane using the magnet’s “natural” B0 variation as an encoding field (SEM)2 and encoding along x is done with Transmit Array Spatial Encoding (TRASE)3. y-z plane resolution depends on the local gradient of the SEM which ideally contains relatively high 1st order fields because blurring results from low-gradient regions of higher order fields4. TRASE requires a spin-echo train of accurate refocusing pulses, which are difficult to achieve over a wide B0 bandwidth (includes built-in SEM)5. Additionally, RF coil quality factors (Q) must be low for wide B0 bandwidths, compromising the coil efficiency. Therefore, the ideal SEM is a tradeoff between resolution and bandwidth. We aim to design a magnet with a large enough bore for human brain imaging, a low B0 bandwidth, and a 1st order field term (specifically X0Y1Z0) that provides at least 2mm in-plane resolution for our standard readout time (6.4ms with SW = 20KHz).

Possible magnet designs are a subset of a full Halbach cylinder model with 3 layers (diameters = 32cm, 41cm, and 50cm) and 24 rungs made up of 1” NdFeB cubes in 18 rows (Figure 1). The magnet bore is smaller than shoulder-width therefore only 7 magnet rows are possible below isocenter (assuming 18cm between shoulders and brain center), so the model includes a 1-row “neck layer” (layer 1) to boost the inferior field. Magnet designs are represented by a vector, X, containing magnet indicator values (0,1, or 2) for each cube location, where 0 indicates air, 1 indicates a grade N42 magnet (Br = 1.32T), and 2 indicates a grade N48 magnet (Br = 1.42T).

We aim to optimize the field in a 20cm DSV at isocenter. The field contribution from each possible cube locations was simulated with Comsol Multiphysics and fit to 6th order polynomials. The field from an arbitrary magnet design is calculated by summing across the appropriate fields specified by the design vector X. MATLAB’s Genetic Algorithm was used to create a high quality population of X vectors that satisfy 2 non-linear constraints: 1) average field > 70mT, 2) X0Y1Z0 coefficient >3.67mT/m. The “fitness” of candidate magnet designs are evaluated by an objective function that removes the X0Y1Z0 field component and outputs the residual resonant frequency range (i.e. minimizes field variation beyond the in-plane first-order term). The algorithm uses the fitness values to determine the next “generation”, and terminates when the best fitness value stops improving in subsequent generations.

Results

Figure 2A shows an example run of the algorithm, and Figure 2B shows properties of resulting solutions from 500 runs of the algorithm. The color indicates the strength of the linear field component. Designs closest to the upper left corner are best because they have high average field strengths (higher SNR) and lower non-linear field variation. Figure 2C shows the same solutions plotted as a function of total field variation (including linear term). The chosen magnet design solution, indicated in Figure 2B and 2C, has an average field of 74mT (3.15 MHz), a range of 2mT (86 KHz) over the target volume, and a X0Y1Z0 coefficient of 4.9mT/m. Figure 3A-B shows the SEM of the non-optimal prototype magnet and the simulated SEM of the optimized magnet. Simulated images using these SEMs are shown in Figure 3C-D, and 1D simulated images of point sources are shown in Figure 4A. Figure 4B shows the width of the resulting point spread functions. Figure 5A shows a drawing of the magnet cube positions (N42=grey, N48=blue). Figure 5B shows a rendering of the planned magnet housing, 32-element receiver-coil array, and head-support piece. The estimated weight of the NdFeB material in this design is 72kg.

Discussion

Imperfections in the NdFeB cubes will cause the constructed magnet’s field to differ from the simulation. For this reason, the magnet housing includes trays that can be populated with shim magnets. After construction, the magnetic field will be mapped and a variation of the described method will be used to place shim magnets.

Conclusion

We have presented a method and resulting design of a lightweight magnet for brain imaging with a built-in SEM. Compared to our previous prototype, this magnet design promises a larger bore, improved resolution, and a total field range that maintains reasonable coil efficiency and transmit bandwidths.

Acknowledgements

The authors thank Elfar Adalsteinsson and Matthieu Sarracanie. Support by NIH R01EB018976, NIH T90DA022759/R90DA023427.

References

(1) Cooley CZ, Stockmann JP, Rosen MS, Wald LL, et al. 3D Imaging in a Portable MRI Scanner using Rotating Spatial Encoding Magnetic Fields and Transmit array spatial encoding (TRASE). In Proc. of the ISMRM, Toronto, Canada, 2015, page 0703.

(2) Cooley CZ, Stockmann JP, Rosen MS, Wald LL, et al. Two-dimensional imaging in a lightweight portable MRI scanner without gradient coils. Magn Reson Med. 2015 Feb;73(2):872-83.

(3) Sharp JC, King SB. MRI using radiofrequency magnetic field phase gradients. Magn Reson Med. 2010 Jan;63(1):151-61.

(4) Schultz G, Hennig J, Zaitsev M, et al. Reconstruction of MRI data encoded with arbitrarily shaped, curvilinear, nonbijective magnetic fields. Magn Reson Med. 2010 Nov;64(5):1390-403.

(5) Sharp JC, King SB, et al. Point-Spread-Functions for RF Imaging with TRASE: Implications for Phase Gradient Coil Design and Flip Angle Calibration. In Proc. of the ISMRM, Stockholm, Sweden, 2010, page 1469.

Figures

Full Halbach cylinder model, containing all possible 1” magnet cube locations and 20cm sphere at isocenter. There are 11 magnet rows above and 7 below isocenter (due to patient’s shoulders). Layer 1 goes around patient’s neck to compensate for asymmetry along x. Layer diameters are 32cm, 41cm, and 50cm.

A) Fitness (penalty) values from one Genetic Algorithm run. B-C) Properties of resulting magnet fields from 500 runs of the algorithm, showing the resonant frequency range with (B) and without (C) the X0Y1Z0 term. The chosen design has a low residual frequency range and a relatively high average field strength.

left column: prototype (non-optimized) magnet, right column: optimized magnet design. (A-B) SEM fieldmaps at x=0cm. (C-D) Simulated images using fieldmaps above and 8 surface coil profiles. A sequence with 180 magnet rotations spaced 2deg apart, and 6.4ms readout (SW=20KHz) was simulated. Images were reconstructed with the preconditioned conjugate gradients method.

A) 1D images of a line of point sources spaced 5mm apart were simulated for the old and new magnet field maps (Fig. 3A-B). B) Full-width at half-max (FWHM) measurements of the point-spread-functions are shown as a function of distance from center of FOV for the simulated SEMs.

A) Chosen NdFeB magnet distribution from solution options in Fig. 2B-C. Blue cubes are grade N48 and grey cubes are grade N42. B) Drawing of magnet housing with helmet for 32 channel array coil. The magnet housing will consists of fiberglass tubes and plastic support and end rings.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3556