Numerically optimized design for a low-cost, lightweight 86mT whole-brain magnet
Patrick C McDaniel1,2, Clarissa Zimmerman Cooley2, Jason P Stockmann2, and Lawrence L Wald2,3

1Massachusetts Institute of Technology, Cambridge, MA, United States, 2Athinoula A Martinos Center for Biomedical Imaging, Charleston, MA, United States, 3Harvard Medical School, Boston, MA, United States


Whole-brain imaging is a major use of MRI, but the cost and siting requirements of scanners limit its use. Here, we propose a close-fitting, lightweight, whole-brain MRI system to address these limitations. We design the B0 magnet for this system using a novel optimization approach and compute simulated B0 maps using 3 magnet modeling approaches. In doing so, we demonstrate the feasibility of realizing an acceptably-uniform whole-brain MRI magnet with mean B0 of 86mT and weighing under 25kg.


Despite its widespread clinical utility, the cost and siting requirements of MRI scanners precludes its use in many applications including point-of-care monitoring and diagnostics. Recently, there has been interest in relaxing these hardware requirements in order to produce portable and low-cost MR systems (1–3). In this work, we address a principal challenge of point-of-care brain MR: designing a lightweight, portable, low-cost magnet with a suitably uniform B0-field over a head-sized ROI for conventional gradient encoding. Magnet homogeneity over the volume of interest is a major determinant of system performance if switched gradient encoding is used. In contrast, an inhomogeneous magnetic field will: require a high readout BW (possibly exceeding that of the RF coils), need high-bandwidth RF pulses, and preclude gradient echoes by introducing excessive intravoxel dephasing. Here, we follow the trend of close-fitting field generation devices initiated in RF receive helmets and propose a close-fitting whole-head MRI helmet-magnet constructed of NdFeB blocks. As we have shown with similar magnets, gradient coils would be positioned external to the B0 magnet and RF coils would be placed inside (Figure 1). We optimize the distribution of rare-earth magnets needed to maximize homogeneity over a brain-shaped ROI and compare 3 different modeling approaches for the NdFeB magnets: a distributed magnetization model, a multipole model, and a dipole model. The results show that a whole brain magnet with B0 = 86mT and weighing approximately 24 kg is possible with acceptable homogeneity over the brain.


Ideal “Halbach magnets” produce spatially-uniform magnetic fields (4,5), but can only be approximated in practice. Previous work optimized truncated cylindrical Halbach magnets for MRI using a genetic algorithm by varying magnet material (6). Here, we use an interior point method to optimize block size (and thus magnetic dipole size) for a helmet-shaped Halbach geometry. We optimize the 3 components of a magnetic dipole moment vector at 296 points on a bulb-shaped surface that surrounds an adult head/neck (Figure 2A) to design a helmet with 296 magnet blocks that minimizes the absolute range of B0 magnitude over a head-shaped ROI (Figure 2B). This ROI matches the geometry of a representative adult head, and included all anatomy above an Axial->Cor plane inferior to the brain. The optimization required a minimum mean B0 of 75mT, and constrained all magnetic dipole moment vector magnitudes be less than that of a 1”x1”x1” block of N52 magnet material. The optimization was performed using Matlab and used the published “test-tube magnet”(7) as an initial guess solution (Figure 2C). In this optimization, each magnet in the assembly was modeled as an ideal point dipole source. Next, each dipole moment vector in the optimized solution was uniformly scaled up until the dipole moment with the largest magnitude matched that of a 1”x1”x1” block of N52-grade NdFeB material. A design was then generated consisting of N=296 non-intersecting N52 magnet blocks of differing volume, such that each block’s magnetic dipole moment matched that generated by the numerical optimization. This model was then evaluated using three numerical tools: the in-house (“dipole model”) code used during optimization; Biot-Savart (Ripplon); and Comsol, and the resulting B0 maps were evaluated. The in-house code models each magnetic block as an ideal point dipole source. Biot-Savart models multipole terms up to fifth order; Comsol accounts for the spatial magnetization distribution and permeability (for N52 NdFeB, μr=1.05) of each block.


The optimized magnetization distribution and physical magnet model are shown in Figure 3. The final design has max linear dimensions of (x=35cm, y=36cm, z=36cm) and, based on the density of N52 material, weighs 24.1kg (not counting the magnet former). Simulated field maps computed using the dipole model, Biot-Savart, and Comsol are shown in Figure 4. The simulated mean B0 values were: 85.8mT, 85.9mT, and 84.3mT as computed with the dipole model, Biot-Savart, and Comsol, respectively. For the three simulations, the corresponding B0 ranges over the ROI were: 0.42mT, 2.4mT, and 2.7mT, respectively. The FWHM of the B0 histograms were 0.042mT, 0.094mT, and 0.11mT, respectively.


Here, we design a magnet for a novel whole-brain MRI system. The magnet was optimized specifically for a head-shaped region, had a mean field of 84.3mT and range of 2.7mT across this ROI, weighs 24.1kg, and is 35x36x36cm in size. Constructing this magnet will involve approximating each dipole moment value as a physically-realizable combination of magnetic blocks of different size and material, which will then be glued into a magnet former. Finally, we will complete the proposed system by designing and constructing the required RF and gradient encoding hardware.


NIH: 5T32EB1680, R01EB018976


1. Cooley CZ, Stockmann JP, Armstrong BD, Sarracanie M, Lev MH, Rosen MS, Wald LL. Two-dimensional imaging in a lightweight portable MRI scanner without gradient coils. Magn. Reson. Med. [Internet] 2015;73:872–83. doi: 10.1002/mrm.25147.

2. Vaughan JT, Wang B, Idiyatullin D, Sohn S, Jang A, BelaBarra L, Garwood M. Progress Toward a Portable MRI System for Human Brain Imaging. In: ISMRM. ; 2016. p. 498.

3. Landeghem M Van, Danieli E, Perlo J, Blümich B, Casanova F. Low-gradient single-sided NMR sensor for one-shot profiling of human skin. J. Magn. Reson. [Internet] 2012;215:74–84. doi: 10.1016/j.jmr.2011.12.010.

4. Halbach K. Design of permanent multipole magnets with oriented rare earth cobalt material. Nucl. Instruments Methods 1980;169:1–10.

5. Leupold H, Potenziani II E. Novel High-Field Permanent-Magnet Flux Sources. IEEE Trans. Magn. 1987;MAG-23:3628–3629.

6. Cooley CZ, Haskell MW, Cauley SF, Sappo C, Lapierre CD, Ha CG, Stockmann JP, Wald LL. Design of Sparse Halbach Magnet Arrays for Portable MRI Using a Genetic Algorithm. IEEE Trans. Magn. 2017.

7. Chen J, Xu C. Design and Analysis of the Novel Test Tube Magnet as a Device for Portable Nuclear Magnetic Resonance. 2007;43:3555–3557.


Figure 1: (A) Concept drawing of system illustrating the sliding magnet/gradient assembly. (B) System drawing with gradient coils removed, showing the head-only magnet and RF coil.

Figure 2: (A) High-field adult head MRI showing definition of magnet surface (blue), isocenter (purple) and target ROI for field design. (B) Target field ROI mask with magnet design surface dimensions. Target point resolution is 7mm isotropic; max linear dimension is 217mm (along z, corresponding to the A-P anatomical axis). (C) 3D target point cloud (orange) and initial-guess discrete magnet design (blue)

Figure 3: (A) Optimized magnet design showing all magnetization vectors. The magnetization geometry approximates a Halbach geometry, but accounts for the unusual, truncated magnet surface. (B) XZ-plane view and (C) isometric view of magnet design generated from optimized design. Magnetized blocks differ in volume, but are all made of N52 permanent magnet material.

Figure 4: (A) Simulated field maps computed using: in-house “dipole model” code (each magnet block modeled as an ideal dipole); a “multipole” model (each block modeled as a sum of 5 multipole terms); and an FEM solver (each block modeled as a magnetization distribution with permeability μr=1.05). (B) Histograms showing the computed B0 distributions across the ROI; ROI minimum and maximum B0 are shown (yellow lines). (C) Table of computed magnet metrics: mean B0; B0 range across the ROI; and B0 distribution full width at half max (FWHM).

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)