Two-dimensional imaging in a lightweight portable MRI scanner without gradient coils

Abstract

Purpose: As the premiere modality for brain imaging, MRI could find wider applicability if lightweight, portable systems were available for siting in unconventional locations such as intensive care units, physician offices, surgical suites, ambulances, emergency rooms, sports facilities, or rural healthcare sites.

Methods: We construct and validate a truly portable (<100 kg) and silent proof-of-concept MRI scanner which replaces conventional gradient encoding with a rotating lightweight cryogen-free, low-field magnet. When rotated about the object, the inhomogeneous field pattern is used as a rotating spatial encoding magnetic field (rSEM) to create generalized projections which encode the iteratively reconstructed two-dimensional (2D) image. Multiple receive channels are used to disambiguate the nonbijective encoding field.

Results: The system is validated with experimental images of 2D test phantoms. Similar to other nonlinear field encoding schemes, the spatial resolution is position dependent with blurring in the center, but is shown to be likely sufficient for many medical applications.

Conclusion: The presented MRI scanner demonstrates the potential for portability by simultaneously relaxing the magnet homogeneity criteria and eliminating the gradient coil. This new architecture and encoding scheme shows convincing proof of concept images that are expected to be further improved with refinement of the calibration and methodology.

Keywords: Halbach magnet; low-field MRI; nonlinear SEMs; parallel imaging; portable MRI.

Figure 1

The magnet array consists of twenty 1”× 1”× 14” NdFeB magnets oriented in the k = 2 Halbach mode. Additional Halbach rings made of 1”× 1”× 1” magnets were added at the ends to reduce field fall off along the cylindrical axis. (A,B) Simulation of the magnetic field in two planes. The field is oriented transverse to the cylinder axis (z-direction). (C) Schematic of NdFeB magnets composing array. The targeted spherical imaging region (18 cm dia.) is depicted at isocenter. (D) End-view photo of the Halbach magnet mounted on high friction rollers. Magnet was constructed with ABS plastic and square fiberglass tubes containing the NdFeB magnets. Faraday cage not shown.

Figure 2

Measured Larmor frequency maps of the spatial encoding magnetic field (SEM) in the z-y (imaging plane), z-x, and y-x planes of shimmed Halbach magnet. The B₀ field is oriented in the z direction.

Figure 3

(A) Linear array of 7 NMR field probes used for mapping the static magnetic field. The probes are held stationary, while the magnet is rotated around them and points on the 2D center plane are sampled. (B) Measured field map for the center transverse slice through the magnet after fitting 6^th order polynomials to the probe data. The black dots mark the location of the probe measurements. The field is plotted in MHz (proton Larmor frequency). This field distribution serves as the SEM information used in image reconstruction.

Figure 4

Schematic depiction of the generalized projections (bottom row) of an object onto the rotating SEM field. The object consists of two water-filled spheres depicted as dashed black lines which are superimposed on the Halbach magnet's SEM field at a few rotations (black arrow depicts B₀ orientation). The NMR spectrum was acquired with a single volume Rx coil.

Figure 5

(A) Photo of the 8 channel receiver array coil with 3D printed disk-phantom at isocenter. The 14cm diameter array is made up of eight, 8 cm loops overlapped to reduce mutual inductance. (B) Relative voxel size is illustrated as a function of radius from the center using two rotations of the magnet's SEM (field isocontour lines illustrated in figure). Symmetry of the isocontours causes aliasing of each voxel through the origin. Using the local sensitivity profiles of an encircling array of coils, the correct location of each signal source in the FOV can be resolved. (adapted from (28)) (C) Photo of the 25 turn, 20cm diameter, 25cm length solenoid transmit coil.

Figure 6

Biot-Savart calculation of the sensitivity map of the Rx coil array. The white arrows show representative orientations of B₀, which define the spin coordinate system orientation (x’,y’,z’). Image reconstruction requires accurate coil sensitivity profiles for each B₀ angle used in the experiment. (A-B) B₁⁻ magnitude and phase for a single representative surface coil located at the right side of the FOV (position marked with white line). Because of the symmetry of the coils’ at isocenter, the coils’ x’ component is approximately zero, and the process of taking the projection onto the x’-y’ plane (to solve for B₁⁻) will produce a vector parallel or anti-parallel to y’. Therefore, the B₁⁻ phase is always +90° or −90° in the depicted transverse isocenter plane. (C) B₁⁻ magnitude of 4 different coils of the array (marked with white lines) for a single magnet rotation position.

Figure 7

Experimental 256×256 voxel, 16cm FOV images of a 3D printed phantom with CuSO₄ doped water occupying the interior of the letters and polycarbonate plastic surrounding it. The phantom has a 13cm dia. and is 1.5cm thick. 91 magnet rotations spaced 2° apart were used, readout bandwidth/Npts = 40 KHz/256, TR = 550ms, spin-echo train length = 6 or 16, with 8ms echo-spacing. Echoes in the spin-echo train for a given rotation were averaged. (A) Image acquired with solenoid Rx coil (32 averages of a 6 spin-echo train). (B) Image acquired with 7 coils of the Rx coil array (8 averages of a 16 spin-echo train). Temperature drift was not corrected for. (C) Image from same data as (B), but with temperature drift correction implemented.

Figure 8

Experimental 256 × 256 voxel, 16cm FOV image of a 1 cm thick slice of lemon placed off axis in the magnet. 5 receiver coils of the array were used to acquire 1 average of a 128 spin-echo train, readout bandwidth/Npts = 40 KHz/256, TR = 4500ms, echo-spacing = 8ms. A) 91 magnet rotations spaced 2° apart were used (B) 181 magnet rotations spaced 1° apart were used.

Figure 9

Simulated images using the calculated sensitivity profiles of the 8 coil Rx array to generate the forward model for 181 1° rotations of the encoding field, 6.4ms, 256 point readouts. The data seen by the Halbach scanner was simulated by processing this “object” through the forward model and adding noise to make it consistent with the SNR of the time-domain signals measured in a water phantom. The model data was then reconstructed using the Algebraic Reconstruction Technique in a 16cm FOV. (A) Reference high resolution 3T T₁ weighted brain image used as the model object. Note: the SEMs were scaled to the brain FOV. (B) Simulated reconstruction using the measured SEM to generate the forward model. (C) Simulated reconstruction using the measured SEM with the additional artificial linear field component (500 Hz/cm). (D) Simulated reconstruction of a 2.5mm grid numerical phantom. Only one quadrant of the FOV is shown, the center of the FOV is marked with white cross-hairs in the upper right.