Arriving at the trailhead

In recent years, there has been renewed and widespread interest in ultra-low-, low-, and mid-field MRI (< 1 T), due to its accessibility, portability, and the potential for unique clinical applications [1,2,3,4,5]. Lower fields offer favorable imaging properties (e.g. reduced SAR, reduced artifacts), and different contrast mechanisms to be explored. Moreover, low-field MRI system performance can be maximized by leveraging recent improvements in MRI hardware and system design, as well as advanced acquisition, reconstruction, and image enhancement methods that utilize readily available computational power. Together, these developments and opportunities make low field MRI a viable and valuable proposition for clinical imaging.

A key motivation to develop low-field MRI technology is enabling new clinical applications. The target applications will define the system design, and can include: portable systems brought to the patient; point-of-care imaging in new healthcare settings outside traditional Radiology enterprises; targeted system geometries, for example for obese patients, exercise imaging, standing imaging, or body-part-specific scanners; field-cycling systems that probe contrast dispersion across field strengths; MRI-guided interventions that leverage reduced metallic device heating; imaging patients with implants; imaging of anatomy with high susceptibility gradients, such as air-tissue interfaces; or even routine imaging offered at a lower cost [6,7,8,9,10].

While most current clinical MRI systems take a “one-size-fits-all” approach, with a single footprint and set of specifications, the variety of low-field MRI systems allow multi-scale imaging with different resolution and imaging properties targeted to answer the clinical question at hand. Low-field MRI is not intended to replace conventional systems or high-field systems, but rather to diversify the MRI use for specific patients, body parts and healthcare environments, similarly to what X-ray based modalities or ultrasound have offered for decades.

This special issue on low-field MRI includes three review articles, one commentary, and twelve research articles providing a representative sample of the ongoing, rapidly-evolving developments in low-field MRI.

Setting out: history, new technology and regulatory approaches

As pointed out by Hennig in his review article, at the beginning of MRI in the late 1970s and early 1980s, all MRI was at low field [11]. The only human-scale magnets that were available operated at low field (even the superconducting magnets), and many people thought that higher fields would not be beneficial in clinical MRI. The situation changed dramatically in 1983 when GE introduced their 1.5 T scanners, leading to a rapid expansion of the “high field” MRI market, and a decline in the use of low field systems. During subsequent decades there have been enormous advances in MRI technology, including parallel imaging with improved RF coils and receiver chain hardware, as well as huge improvements in gradient hardware, leading to significant gains in achievable SNR and sampling efficiency. These, together with the steady development of advanced image reconstruction methods, including AI-based methods, to make best use of the available SNR, have enabled a revival of low-field (and even ultra-low-field) MRI, which is once again viable for clinical diagnosis.

With several companies now producing low-field and point-of-care MRI systems, regulatory authorities around the world must decide how to assess the “new” imaging technology (even though low-field scanners were approved for medical use decades ago). Now, the regulatory bodies must take into consideration not only the scanners’ field strength, but also the recently-introduced technologies including the use of AI. In their review article, Krainak et al. provide a detailed description of the US regulatory framework and discuss how the Food and Drug Administration evaluates low-field MRI systems that have been submitted for regulatory approval [12]. Other countries’ regulatory authorities will inevitably adopt their own procedures, but the general approaches are likely to be similar.

Packing light: low-field system design and hardware

There are several approaches to design an ultra-low-, low- or mid-field system with increased design flexibility in the lower field regime. Often, the target applications inform the system design. Recently, to evaluate whole-body imaging opportunities at mid field, researchers have ramped down existing clinical scanners, e.g. to 0.55 T. Guenthner et al. used a similar ramp-down strategy, but also leveraged existing multi-nuclei hardware to temporarily convert a 3 T scanner to 0.75 T [13]. The ramped down field strength of 0.75 T was cleverly chosen to match the proton Larmor frequency to the existing 3 T C13 RF hardware (32 MHz). Therefore, instead of replacing the RF subsystem, the existing C13 coils and spectrometer were used, allowing an inexpensive and efficient temporary system conversion to 0.75 T. Cardiac and abdominal imaging and MR spectroscopy results from the ramped down system are presented from that system.

Moving from fixed conventional scanners to portable point-of-care scanners leads to a giant leap in accessibility but also a giant gap in signal-to-noise ratios. As SNR decreases roughly linearly with B0, truly diagnostic image quality is challenging for portable scanners operating in the < 100 mT range. Therefore, every step in the system design process should focus on maximizing SNR within the predefined geometrical constraints for the scanner. Through this lens, Webb et al. review the major system components and design considerations for low-field point-of-care MRI scanners [14]. The designs and tradeoffs in permanent magnets are reviewed, including C- and H-shaped yoked magnets and Halbach arrays. A detailed analysis of RF coils for low field is included, with unique considerations for the performance (i.e. Q values) and ultimately SNR. Also reviewed is the design process for the remaining RF subsystem, console, and methods for decreasing interference and noise. Finally, a summary of point-of-care systems in the literature is reviewed.

The article by de Vos et al. describes a framework for the design of such low field point-of-care scanners based on cylindrical Halbach magnets [15]. The framework uses computational methods in an interactive loop to design the magnet, gradient and RF coils, and allows the designer to explore the interdependency of the components and imaging specifications. The approach employs user inputs for imaging requirements (resolution, point-spread-function, volume), RF and gradient parameters, and magnet parameters, and iterates through system designs with the aim of minimizing the system size. Each system component (magnet, gradient coils, and RF coils) is optimized with a target field approach with considerations for coil efficiencies. To demonstrate the framework, a simulated extremity scanner and neuroimaging scanner are presented.

In contrast to the cylindrical Halbach magnet, Wei et al. describe the design of a moveable yoked H-shaped permanent magnet for brain imaging [16]. At approximately 0.2 T, this new magnet design produces the highest field strength so far in a portable head scanner, and approaches the well-established field strengths of conventional permanent magnet open-bore scanners. Recently published portable H-shaped magnet systems have been at approximately 0.05–0.065 T and weight of 600–750 kg; this new design offers a dramatic increase in field strength while maintaining a similar weight (654 kg) and an impressive homogeneity of 46 ppm. This gain in efficiency is partially afforded by a smaller magnet gap and imaging volume and potentially less robust rare earth material. While there are clear tradeoffs in these design decisions, it is exciting to see a movable MRI scanner approaching widely accepted field strengths.

In addition to magnet design, RF receiver coil optimization for specialized applications is critical for maximizing the SNR and ultimate clinical potential of low-field MRI. Munoz et al. presents a dedicated Rx array for high-performance 0.55 T dynamic speech imaging [17]. This new Rx array is specially designed for upper airway MRI at 0.55 T and evaluated at 11 defined upper airways volumes of interest. The g-factor and SNR of the 0.55 T dedicated coil clearly outperforms both the integrated body coil and a 16-channel head-neck coil, motivating the need for these types of specialized coil designs.

Navigating the terrain: MRI safety

Despite common belief, a lower field strength does not always equate to lower MR risk. MRI safety is multifactorial, and blanket safety statements labeling low field-systems as “safer” may prove misleading for specific clinical scenarios, devices, and patients. In their commentary, Gilk and Kanal describe the benefits of reduced torque, lesser spatial field gradients, and lower SAR at lower fields [18]. However, they point out that torque may be applied in a different direction on an implant for magnets with a different B0 orientation, and that RF heating is dependent on the resonant wavelength, and while wavelengths are longer at lower field, this does not preclude a safety risk for certain long implants and devices. Importantly, they point out that, to-date, most implants are not labelled for field strengths < 1.5 T.

In this special issue, two research papers address RF safety for coil designs and interventional devices. Parsa and Webb present comprehensive electromagnetic field simulation results for four field strengths ranging from 0.05 T to 0.1 T for commonly used coil geometries [19]. Their simulation studies confirm SAR-efficiency was higher and that SAR was not a significant issue for emerging point-of-care MRI systems. Ozen et al. systematically investigates the RF heating of several interventional devices such as guidewires, catheters, biopsy needles and a microwave applicator, intended to be used for MRI-guided interventional procedures at 0.55 T [20]. Device heating will depend on the device length, insulation, patient size, and intended device trajectory for each application. They performed measurements of the electric field, temperature, and transfer function, which did not lead to any significant heating, supporting the use of low field MRI for MRI-guided interventional procedures.

Forging a path: image acquisition and reconstruction methods

The successful implementation of low field MRI relies on the design and optimization of image acquisition and reconstruction strategies, within the constraints of the system design, to deliver the best possible image quality for the specific clinical target. Many of the advanced reconstruction strategies developed at higher fields can have direct and immediate impact when applied at low field, in addition to new developments tailored for low field strength.

Two research articles on image acquisition and reconstruction are included in the special issue, providing examples of these techniques in action. Hamilton et al. describe the development of a novel self-supervised low-rank deep image priors image reconstruction method applied to real-time (free-breathing and ungated) spiral cardiac cine acquisition for a commercial 0.55 T system [21]. The method combines two U-Nets to generate spatial and temporal basis functions to compactly represent dynamic images, and the authors demonstrate the feasibility of this technique in healthy volunteers. Meanwhile, Ramasawmy et al. describe an approach to mitigate concomitant field artifacts in highly efficient spiral turbo-spin-echo neuroimaging [22]. Concomitant fields generate unintentional spatially varying phase that scales inversely with field strength (i.e. 1/B0) and quadratically with gradient amplitude. The authors’ approach involves the addition of bipolar gradient waveforms before and after each readout to mitigate phase accumulation at each echo time and between echo spacings. They specifically illustrate the impact of phase errors through simulation and demonstrate their technique for T2-weighted neuroimaging in healthy volunteers at 0.55 T.

A steep, yet rewarding climb: quantitative MRI

A key aspect to guarantee deployment of new technologies, and hence clinical adoption, is one’s ability to provide reliable diagnosis, leveraging reproducible methods while using robust tools. Quantitative imaging, although particularly difficult to implement in low-SNR regimes, has the advantage to provide reference values not only for quality control and quality assessment but also for longitudinal studies during treatment follow-up and/or comparison across patient populations. Four research papers in this special issue tackle this challenge, two of them using conventional imaging, and the other two using fast acquisition strategies.

Martin et al. present temperature controlled T1 and T2 relaxation times in calibrated phantoms at different magnetic field strengths (0.0065, 0.0640, and 0.5500 T) which could serve as benchmarks for quality assessment of low-field MRI systems and methods [23]. Jordanova et al. conducted a study at 64 mT in healthy volunteers where brain T1 and T2 values are reported using 6- and 10-point inversion recovery and spin echo sequences, respectively [24]. Reference values (means and standard deviations) were computed using automatic segmentation in three main compartments: white matter, grey matter, and cerebrospinal fluid. O’Reilly and colleagues assessed MR Fingerprinting (MRF) performances in muscles and lipids at 50 mT using a custom optimization scheme tuned to specific pairs of T1/T2 values [25]. In combination with matrix completion, they could achieve in vivo volumetric mapping in 13 min. Finally, Liu et al. compared the performances of spiral and rosette MRF for single-slice, simultaneous T1, T2 mapping at 0.55 T while taking advantage of the inherent off-resonance signal suppression property of rosette trajectories to enable water-fat separation [26]. The sequences (≤ 15 s) were tested in phantoms and proof-tested in the livers of 16 healthy controls.

A new perspective on the scenery

The papers of this special issue, as well as other recent low-field MRI works published in the literature, clearly show that low-field MRI is no longer considered a niche which aims solely at delivering low-quality images using low-cost technologies. Instead, the community’s effort to deliver always more robust and advanced methods for low-field MRI shows that it strives to provide high-quality standards for reliable diagnosis that ultimately must meet clinical needs.

After effort comes comfort: let us all enjoy a new day on the foothills of MR imaging. Exploring foothills is certainly not as trivial as it sounds, and the challenges found on the path will lead to great rewards. This special issue sheds light on where the community stands in mid-2023, when it comes to low-field MRI systems and methods but does not yet reflect where we stand in terms of applications. More is on its way to demonstrate the role of low-field MRI for neuroimaging applications (such as in intensive care units or acutely in patients with stroke) [27, 28], but also body (abdominal, lung, musculoskeletal imaging) [29] and cardiac MRI (diagnostic and/or interventional) [30]. The potential of lower field strengths does not stop at cost considerations, but the physics of MRI indicates the rise of a new realm of applications and contrasts, diversifying the scope of MRI diagnostics to places and settings so far barely accessible. The major promise lower field strengths offer stands in their versatility, and are yet to be fully exploited—be prepared to continue the journey in the foothills of MRI!