A portable scanner for magnetic resonance imaging of the brain

Abstract

Access to scanners for magnetic resonance imaging (MRI) is typically limited by cost and by infrastructure requirements. Here, we report the design and testing of a portable prototype scanner for brain MRI that uses a compact and lightweight permanent rare-earth magnet with a built-in readout field gradient. The 122-kg low-field (80 mT) magnet has a Halbach cylinder design that results in a minimal stray field and requires neither cryogenics nor external power. The built-in magnetic field gradient reduces the reliance on high-power gradient drivers, lowering the overall requirements for power and cooling, and reducing acoustic noise. Imperfections in the encoding fields are mitigated with a generalized iterative image reconstruction technique that leverages previous characterization of the field patterns. In healthy adult volunteers, the scanner can generate T1-weighted, T2-weighted and proton density-weighted brain images with a spatial resolution of 2.2 × 1.3 × 6.8 mm³. Future versions of the scanner could improve the accessibility of brain MRI at the point of care, particularly for critically ill patients.

Fig. 1 |. Portable MRI brain scanner prototype.

a, The scanner main components are inside the 56 cm diameter magnet (orange cylinder). The amplifiers console and computer are not shown. The subject’s shoulders remain outside the magnet, allowing for a lightweight, small bore design that fits the head only. The patient table detaches from the scanner cart to facilitate transport. b, Exploded computer-aided design (CAD) model of the main scanner components (from left to right): spiral transmit/receive RF helmet coil, Halbach magnet cylinder, 2-axis gradient coil, and RF shield. c, Corresponding photo of exploded view.

Fig. 2 |. Permanent low-field magnet design.

a, The B₀ = 80 mT cylindrical Halbach magnet has an outer diameter = 56 cm, length = 48 cm, total weight = 122 kg (80 kg of rare earth material). Photo shows the superior side (“service end”) of the magnet with a 35.3 cm diameter opening. The inferior side of the magnet (shoulder side) has a 27 cm bore opening due to the 32 cm dia. ring of 1” “booster” magnets near the shoulders placed to alleviate the field fall-off. b, Close-up photo of superior end of magnet. The 1” NdFeB magnets are contained within the square cross-section fiberglass tubes. The two main magnet layers are at radii 20.5 cm and 25 cm. The white plastic shim trays contain the addition of smaller NdFeB magnets to further optimize the magnet field. c, CAD model showing the distribution of N52 grade (grey) and N42 grade (white) NdFeB 1” cubes comprising the Halbach magnet optimized for a built-in monotonic read-out encoding field in the x direction. d, Measured field-map in the axial 18 × 20 cm planes for the constructed magnet distribution prior to shimming. The 17 × 14 cm ovals outline approximate brain dimensions. e, CAD model of shim magnet distribution for fine-tuning of the field. The smaller shim magnets axial position was fixed, but size (< 1/4” cube) and the dipole direction were varied. F, Measured field-map of the shimmed magnet, showing an improvement in the field linearity in x.

Fig. 3 |. Gradient coil design.

a, G_y and G_z gradient coils with wires press-fit into a tiered cylinder 3D printed former. The G_y gradient coil is on the outer surface and G_z is on the inner surface. The tiered shape allows for maximum diameter (34.8 cm) and length (42.7 cm) within the magnet. b-c, Gradient coils’ current density contours designed with a BEM stream function method optimized for linearity in the 20 cm ROI. d-e, The measured gradient coil field maps for 1 A of drive current in the coils. The G_y and G_z coil efficiencies were 0.6 mT/m/A and 0.8 mT/m/A respectively.

Fig. 4 |. MRI pulse sequence diagram.

The 3D RARE (Rapid Imaging with Refocused Echoes) pulse sequence is shown for proton density (PD) weighted sequence. The RF applies the 90-degree excitation chirped pulse (3.2 ms, 100 kHz sweep) followed by a train of 180 degree chirped refocusing pulses (1.6 ms, 100 kHz sweep). The phase of the pulses follows a phase cycling scheme that prevents mixing of the resulting FID and Spectral echoes. The G_x readout gradient is the built-in permanent magnet encoding field, and therefore is continuously applied throughout the acquisition. The G_y gradient produces phase encoding blips that vary along the echo train for partitioning data in the y dimension completing the 23 encodes in each shot. The G_z phase encoding blips are incremented shot-to-shot requiring 97 TR periods to complete the encoding. The Signal Acquisition alternates between the narrow “FID echoes” and wider “spectral echoes”. The sequence is converted to T₁-weighting with the addition of an initial inversion pulse. In the T₂-weighted sequence, the ordering of the G_y phase encoding blips are re-arranged so that the center of k-space is captured at TE_eff = 167 ms.

Fig. 5 |. 3D T2, T1 and PD-weighted images of the brain in healthy adult volunteers.

A subset of the acquired 23 partitions are shown. Image resolution ~ 2.2 × 1.3 × 6.8 mm³. The first 5 rows show images reconstructed with the generalized forward-model based reconstruction method. S1 PD: (subject 1, male 63 years old) PD images acquired with 3D RARE, TR/TE_eff = 2900ms / 14ms, acquisition time = 9:24 min (2 averages). S1 T1: (subject 1) T₁-weighted images acquired with inversion prepped 3D RARE, TI/TR/TE_eff = 400ms / 1830ms / 14ms, acquisition time = 11:46 min (4 averages). S1 T2: (subject 1) T₂-weighted images acquired with 3D RARE sequence, TR/TE_eff =3000ms/167ms, acquisition time = 9:42 min (2 averages). S2 T2: (subject 2, male 63 years old) T₂-weighted images acquired with 3D RARE sequence, TR/TE_eff =3000ms/167ms, acquisition time = 9:42 min (2 averages). S3 T2: (subject 3, female, 53 years old) T₂-weighted images acquired with 3D RARE sequence, TR/TE_eff =3000ms/167ms, acquisition time = 19:24 min (4 averages). S3 FFT: the S3 T2 data reconstructed with a conventional FFT reconstruction instead of the generalized method. This last image demonstrates the geometric distortion that results from the non-linear encoding fields when the field-maps are not included in the reconstruction model. The measured SNR in the images were SNR = 127, 80, 68, 65, 124 for the image acquisitions in rows 1–5 respectively.

Fig. 6 |. Analysis of the measured encoding field-maps (G_x, G_z, and G_y) in the central field-map slices.

The ideal maps are calculated as a linear fit to the measured maps. The error maps show the percent difference between the measured maps and ideal maps. The color range is higher (up to 50%) for the G_x gradient. Spatial deformation maps show the resulting image distortion that occurs when the ideal linear map is assumed (instead of the measured map). The non-linearities and spatial deformation are most severe in the G_x encoding map, which is generated by the built-in permanent magnet gradient. The G_x analysis shows high errors near the periphery and severe spatial deformation approaching signal singularities in some locations. In contrast, the gradient coil maps (G_z and G_y) errors and spatial deformation maps are more benign.