Human brain image captured at 10.5 T. Credit: Kamil Uğurbil, Russell Lagore, Andrea Grant and the CMRR 10.5T Team

For many neuroscience researchers, studying the living brain at submillimeter resolution is just another day in the lab. Indeed, many groups have demonstrated the feasibility of imaging neuronal behavior in live animals using optical techniques such as multiphoton microscopy. But the challenges mount quickly when the goal is to access the full brain — and in humans, the invasive procedures routinely used in rodents or primates are impractical or unacceptable.

This helps explain the enthusiasm surrounding the recent release of the first human brain imaging data from Iseult1. This Paris-Saclay, France-based magnetic resonance imaging (MRI) facility is employing a 11.7 tesla (T) magnet, making it the world’s most powerful MRI system for neuroscience research in people. Magnetic field strength is a critical determinant of the resolution, contrast and imaging speed that can be achieved with MRI, and the results from this initial demonstration were encouraging. In roughly five minutes, the Iseult MRI could image the brain at a resolution of 0.19 × 0.19 × 1 mm3. “What you can see in those images is just incredible,” says Noam Shemesh, an MRI scientist at the Champalimaud Foundation in Lisbon, Portugal, who was a reviewer for the Iseult team’s manuscript. “You see resolution that you just have never seen before.”

Researchers at the Iseult 11.7-T ultra-high-field MRI in Paris-Saclay recently published their first set of human brain images. Credit: Nicolas Boulant, CEA

Although such resolution is insufficient for mapping individual cells, this and other ultra-high-field MRI systems unlock the potential to access so-called ‘mesoscopic’ detail in the brain, a critical asset when the goal is to understand this organ’s workings. “When we get to submillimeter resolution … we can see groups of neurons that are behaving as a unit in a neural circuit,” explains David Feinberg, a physicist at the University of California at Berkeley. This makes it possible to study communication between disparate brain structures — even across considerable distances — and ascertain the role of these interconnected circuits in brain function. The resulting insights could explain the core brain features underlying a wide range of behaviors and cognitive processes, and yield mechanistic insights into conditions like epilepsy, Alzheimer’s disease or stroke.

The threshold for ultra-high-field is 7 T, and MRI experiments at this field strength have become increasingly commonplace over the past 20 years. But there is also a steady push for ever-stronger magnets, such as the Iseult facility’s. “We feel like explorers,” says Nicolas Boulant, a physicist at the French Atomic Energy Commission’s NeuroSpin facility and coordinator for the Iseult effort. “It’s like a telescope that you put in space and you’re not really sure what you’re going to see.” But bigger fields also bring bigger challenges, and accessing the benefits of ultra-high-field MRI requires a combination of imaging expertise, scientific savvy and clever engineering.

Reading between the layers

The fundamental principle of MRI entails placing a subject in an opening, known as a ‘bore’, within a powerful magnet. This causes hydrogen nuclei within the bore to align themselves along the orientation of the magnetic field. These nuclei are then transiently bombarded with radiofrequency (RF) pulses from transmitter coils in the device, which are tuned to a specific ‘Larmor frequency’ that can knock the nuclei out of alignment at a given field strength.

Karin Markenroth Bloch, a physicist who heads the Swedish National 7T facility at Lund University in Sweden, compares the mechanism to a child’s swing set. The magnetically induced alignment is akin to pulling the swing back and holding it. “You have to start pushing it for something interesting to happen, and then you have to push it in resonance with the swing,” she says. “If you stop, it will swing back to equilibrium.” In MRI, the restoration of nuclei to their original orientation produces another burst of RF signal, which is detected by the instrument’s RF coils and analyzed to generate an image. In addition to the primary ‘static’ magnetic field, MRI also employs dynamic magnetic fields that are applied as a gradient and make it possible to spatially map each signal.

These data can generate detailed structural images of the brain and other tissues based on the signal produced by hydrogen atoms in molecules including water and lipids. But there are also many other MRI-based imaging methods, such as functional MRI (fMRI), which is widely used by neuroscientists to link specific behaviors and neurological processes with patterns of neuronal activation in the brain. Here, oxygenation of hemoglobin in the blood alters the magnetic properties of this molecule, creating a signal that can be used to determine what areas of the brain are receiving more oxygenated blood, providing an indirect but useful indicator of local neuronal activity.

Most clinical and research MRI systems operate at 1.5 T or 3 T. These systems are extremely useful — for example, the Human Connectome Project relied heavily on fMRI data from 3-T MRI instruments, which allowed the consortium to identify 180 distinct functional regions in each hemisphere of the human cerebral cortex2. But fMRI with standard instruments tends to produce signal ‘blobs’ with resolution on the order of several millimeters, making it impossible to zoom in on the cortical layers and columns that represent the fundamental organizational units of the cerebrum. “The local circuits are in the columns, and all the information flow within and between regions occurs across the layers,” says Natalia Petridou, a neuroscientist at University Medical Center Utrecht in the Netherlands. “It’s the essential structure where any sort of computation and information processing takes place.”

Comparison of human brain images at 7 T and 3 T, showing that a 7-T magnet provides a factor of >160-fold more spatial detail. BOLD, blood oxygen level-dependent contrast. Credit: Natalia Petridou

Capturing this level of detail requires voxels with dimensions on the order of hundreds of micrometers — and this requires a stronger magnet. “You get a gain in signal-to-noise ratio with a higher field strength,” says Markenroth Bloch, adding that this greater sensitivity can be leveraged in a variety of different ways. “If you go from 3 T to 7 T, in the same scan time you can get a higher-resolution image, or maybe you choose to obtain the same image resolution in a shorter time.” One study indicates that every doubling of the strength of the static magnetic field yields a threefold improvement in sensitivity3.

Reaching for resolution

Functional imaging in particular was a critical driver behind the invention of the first ultra-high-field MRI systems. “Functional imaging signals are very weak, and sometimes people try to extract too much out of that very weak signal and end up with crazy and non-reproducible ideas,” says Kamil Uğurbil, a physicist at the University of Minnesota who was among the inventors of fMRI4. “We introduced ultra-high-field because of fMRI, really.”

The world’s first ultra-high-field system, an 8-T MRI, was developed by a team led by Pierre-Marie Robitaille at Ohio State University in 1999. That same year, Uğurbil and collaborator David Rayner constructed the first 7-T system in Minneapolis, and 7 T has subsequently become the standard magnetic strength for most ultra-high-field human imaging studies. But early-generation instruments only teased the potential of 7-T MRI. This is because operating at these ultra-high-field regimes creates a host of new challenges, requiring substantial technological development.

One pressing issue arises from the higher Larmor frequencies required for ultra-high field. This results in wavelengths for the radiofrequency pulses that are smaller than the human brain. “These waves will start interfering with each other, so the tissue is not evenly excited over the whole brain and you will have areas of signal dropout,” says Markenroth Bloch. Today, this is typically remedied through the use of ‘parallel transmission’ techniques, in which the 7-T MRI system is equipped with eight RF coils that can be used to shape the RF pulses for even coverage across the brain.

The increased sensitivity and spatial resolution at 7 T also make it more vulnerable to motion artifacts. For example, Markenroth Bloch says that users might want to image a brain structure like the hippocampus with a spatial resolution of 400 micrometers. “But most people will move more than that — and actually, the hippocampus can move more than that just because of your heartbeat.” Such artifacts remain an ongoing challenge, but parallel progress in software development and image processing is helping to preserve the resolution gains enabled by 7 T. “We work on image reconstruction techniques that actually reduce artifacts that come from heartbeat, breathing and motion,” says Uğurbil, noting that deep learning-based tools have shown particular promise in image reconstruction and are expected to be important in reducing motion artifacts as well. His team also developed a popular denoising tool called NORDIC in 2021, which enables isotropic 0.5-micrometer-resolution fMRI imaging in the brain at 7 T (ref. 5).

Today, there are multiple commercial 7-T instruments on the market, including a system from Siemens that was cleared for clinical use in the United States and Europe in 2017. This has made the technology more accessible, and Boulant believes there are roughly 100 sites with 7-T MRI worldwide.

That said, much of the research conducted at 7 T has focused on demonstrating the feasibility of addressing structural and functional questions with sufficient resolution and fidelity. “Once you can do that, you can start asking biological questions,” says Petridou. “So it’s really starting now, basically.” Still, many promising use-cases have already begun to emerge, particularly in other domains like structural brain imaging. In Lund, for example, Bloch says that roughly one-third of her facility’s 7-T users involve the use of fMRI to pursue clinical research questions related to psychiatric conditions like depression and anxiety and neurological disorders like Alzheimer’s. And in her own research, she is using the instrument to investigate physiological processes that contribute to the drainage of waste products from the brain via the so-called glymphatic system.

Such great heights

Even as the MRI community continues to optimize the performance of 7-T instruments, there has been an ongoing push for still-higher field strengths. This requires a massive infrastructure investment and an army of experts in diverse domains. For example, the University of Minnesota’s Center for Magnetic Resonance Research currently houses America’s most powerful human MRI, a 10.5-T system, in a dedicated facility that cost $53 million to construct — and the magnet alone cost a further $14 million.

The University of Minnesota’s Center for Magnetic Resonance Research, directed by Kamil Uğurbil, houses America’s most powerful human MRI, a 10.5-T system. Credit: Kamil Uğurbil

Many considerations go into magnet design, particularly at this scale. For example, Uğurbil highlights the size of the bore into which subjects are inserted for imaging: a larger bore requires a bigger and more costly magnet, whereas a too-small bore can create interactions with the gradient field. “That interaction can kill the magnet,” he says. This decision is also affected by whether the MRI is capable of whole-body scans, like the Minnesota 10.5-T and Iseult 11.7-T systems, or solely designed for brain imaging, like the pair of 11.7-T systems now under development at the Gachon University Gil Medical Center in South Korea and at the US National Institutes of Health. The latter option can simplify operation more generally, as the inhomogeneities in magnetic field gradients that already confound brain imaging become more severe in larger body parts like the torso. That said, Uğurbil reports that their 10.5-T system has been successfully used to image organs including the heart and liver by employing a parallel transmission system with 16 RF coils.

Most current magnets are based on niobium-titanium superconductors, comprising hundreds of kilometers of tightly coiled wire, and must be carefully managed to operate smoothly. Cooling is essential, and both the Iseult and Minnesota systems use substantial quantities — 7,000 liters for Iseult — of liquid helium to reach operating temperatures below 3 K. Given the scarcity and cost of helium, this substance must be carefully managed and recirculated. The magnet must also be well shielded to prevent it from acting on instruments and other metal apparatus outside the MRI system. The Minnesota system sits within a literal vault of 600 tons of iron, whereas Iseult uses an ‘active shielding’ approach in which an additional magnet surrounds and confines the field from the 11.7-T magnet.

Many of the challenges that emerge at 7 T are further amplified at greater field strengths. The resolution boost makes imaging far more susceptible to blurring, and Boulant notes that they had to toss data from nearly two-thirds of their first 20 human subjects because of motion artifacts. Other sources of noise become worse as well. “You have more vibrations … you have more acoustic noise, and so it’s harder,” says Boulant. “It spans a wide range of different problems, and I don’t think a single lab in the world has all the skills to cover all the problems in a reasonable timeframe.” To this end, the NeuroSpin team is working as part of the European Accurate, Reliable and Optimized Functional Magnetic Resonance Imaging (AROMA) consortium to assemble the necessary expertise.

Despite the challenges, the opportunities are also becoming clear, and the initial Iseult human data have impressed the experts. “It was so exciting to see these gorgeous images with greater contrast between the gray and white matter and higher signal to noise,” says Feinberg. “You’d have to spend four times as long to get those images at 7 T.”

The Minnesota facility has been imaging people since 2018, with 66 human subjects scanned over the course of 166 sessions, and Uğurbil says that his team can now routinely achieve localized resolution of 350 micrometers and whole-brain resolution of 650-micrometer isotropic voxels. This resolution can unlock entirely new realms of MRI-based neuroscience. “If you can see a detailed view of cortical layers and not only two to three pixels per layer,” says Shemesh, “that opens up the way to characterizing the brain in ways that are only possible today when you cut the brain out and put it under a microscope.”

There is also considerable excitement about so-called ‘x-nuclei’ methods, in which magnetic resonance is used for the spectroscopic detection of nuclei other than hydrogen. “We did sodium, we did phosphorus, and carbon-13, and that’s certainly much better than 3 T and also much better than 7 T,” says Klaus Scheffler, a physicist at the Max Planck Institute for Biological Cybernetics in Tübingen, Germany, who works extensively with a 9.4-T MRI system. Ultra-high-field regimes also enable the use of deuterium labeling to image more complex biomolecules. For example, Scheffler’s team is exploring the imaging of deuterated glucose at 9.4 T as a means for characterizing tumor metabolism, and Uğurbil and colleagues even envision opportunities to map the activity mediated by neurotransmitters such as GABA and glutamate. At higher fields, says Uğurbil, “the peaks of this frequency spectrum of different molecules go further apart … so they are better resolved, and of course they have a higher signal.” In the future, this could offer a complementary and more direct view of active brain function relative to conventional fMRI.

Animal advances

The opportunities that emerge at this scale are no surprise to researchers working with animal models, who have been operating at double-digit field strengths for some time. “What is considered ‘high-field’ in humans, we consider low-field almost in rodents,” says Shemesh. His lab will soon be installing the first commercial 18-T instrument, manufactured by Bruker for rodent MRI, and the world’s most powerful MRI system is a system for animal imaging at the National High Magnetic Field Laboratory in Tallahassee, Florida, which operates at a whopping 21.1 T.

This disparity between animal and human MRI capabilities is largely a matter of cost — building a powerful magnet with a rat-sized bore is a far less ambitious undertaking than building one that can accommodate a grown person. Furthermore, issues like inhomogeneities in magnetic field gradients are less severe at the scale of the rodent brain, even as RF frequencies climbs higher. Shemesh acknowledges that for functional imaging, “it’s still not really ‘plug and play’,” but says that structural brain imaging experiments in animals are relatively straightforward even at these extremes of field strength.

These instruments can be powerful tools in the hands of experienced users. For example, Shemesh says that to explore questions about brain development and function, one would normally have to dissect the brains of large numbers of animals to draw statistically sound conclusions. “It’s much more dramatic when you can actually follow the same animal over time and see how it changes and correlate that with all kinds of behavioral measures and potentially do a lot of validations that you cannot do in humans,” he says. Furthermore, his team is already performing some of the MRI-based spectroscopic imaging experiments — for example, detecting neurotransmitters in the living rodent brain6 — that are still in early development stage for humans.

An ultra-high-resolution 3D-rendered image of a mouse brain after stroke acquired at ultra-high field of 16.4 T reveals details including single clots (enlarged, dark contrast) in the stroke-affected area. Credit: Noam Shemesh and Rita Alves

Xin Yu, a biophysicist at Massachusetts General Hospital in Boston, has conducted many animal studies at 14 T and is also enthusiastic about the opportunities that have emerged for preclinical brain research. “20 years ago, I needed two hours at 7 T to achieve 100-micron isotropic imaging in anesthetized mice,” says Yu. “Today, we only need two seconds to get comparable spatial images, and we can study brain function in the awake, behaving mouse.” Yu’s group has been developing methods for discriminating the signal contributions of individual blood vessels in fMRI experiments7, enabling more precise fMRI-based analysis of brain activation, and is also conducting multimodal studies that combine fMRI with methods like fiber photometry. For example, his team is using this approach to simultaneously monitor release of the neurotransmitter acetylcholine in conjunction with standard fMRI imaging of the neurovascular response to explore the interactions between these two processes in both healthy mice and animal models of Alzheimer’s disease.

Not everything is simpler with animals, however. As noted above, movement artifacts become especially severe at ultra-high-field regimes, and being enclosed in a narrow, extremely noisy aperture can make rodents understandably anxious. “We’re talking about easily 100, 120 or 150 decibels,” says Yu. Both he and Shemesh routinely perform several weeks of training to acclimatize animals to remain still in the MRI for an hour or so at a time. In conjunction with tools for immobilizing the animal’s head, this training can mitigate — but not eliminate — the movement problem, and the field is still working toward robust software tools that can eliminate motion-induced artifacts. Sedation or anesthesia are also options, but these undermine efforts to document normal behavior and brain activity, and Yu says that even with careful training, there is evidence that the stress associated with MRI imaging may have a meaningful effect on imaging results.

At the frontiers of ultra-high field

The limits of field strength may be marching upward, but some researchers still see fertile ground to cultivate in the 7-T domain. For example, Feinberg is eager to push the resolution of human fMRI to new heights, but also concerned about making sure these gains are broadly accessible. He notes that “the magnets get very expensive and they’re at high risk” of failing beyond 7 T. “I wanted to build a stable scanner.”

He therefore set about designing a system to push 7 T beyond its current limits, working with partners in both academia and industry. In a 2023 publication8, Feinberg and colleagues presented their ‘NexGen’ 7-T system, which can resolve individual cortical layers with spatial resolution on the order of 350–450 micrometers. Among the key features of this platform are a powerful new gradient coil and state-of-the-art RF receiver array, which allows the instrument to quickly pick up fleeting magnetic resonance signals before they decay. “For fiber tract imaging in neuroscience, we’re now able to get twice the volumetric spatial resolution with about 20% higher signal in the middle of the brain compared to any other 7-T scanner — much more than that, actually,” says Feinberg, adding that the system also maintains higher signal gains at the periphery of the brain.

Importantly, existing commercial 7-T systems can be upgraded to enable NexGen imaging, and Feinberg says at least four sites around the world plan to acquire or have already adopted the system, including the Otto von Guericke University Magdeburg in Germany. “I already did some measurements there, and it’s a fantastic system,” says Scheffler. “Adding this hardware produces close to two times more signal as with the conventional 7 T, depending on the application.” Feinberg believes advances like this are more likely to fuel progress in mesoscale brain imaging in the near term until higher-field systems become sufficiently mature and available for routine experimentation. As a bonus, the technologies developed for NexGen could potentially be applied at 10 T or higher to further amplify the resolution gains delivered by those systems.

But the ultra-high-field arms race is far from over. In the Netherlands, a consortium known as DYNAMIC, led by David Norris at Radboud University, is building the world’s first 14-T MRI for human brain imaging9. “We do have questions where, actually, the problem is not at 500 microns like I thought, it's actually at 200 microns,” says Petridou, who is part of the DYNAMIC consortium. “To see that, I need a stronger magnet.” A similar effort is underway in Germany, spearheaded by Mark Ladd of the German Cancer Research Center, although this initiative is not yet formally funded.

This takes MRI into uncharted territory. The magnet materials currently used for ultra-high-field human MRI have a field strength limit of 12 T. In order to achieve 14 T, the DYNAMIC team will therefore be employing a first-of-its-kind magnet based on a high-temperature semiconductor material developed by Magdeburg-based MRI company Neoscan Solutions. “It’s a certain risk — the magnet has never been built,” says Scheffler, who is part of the German 14-T initiative and is also collaborating with DYNAMIC. “It’s kind of a small and new company, but they have very competent people.” Other major challenges emerge at this scale. Ladd cites the forces that come into play due to the extreme magnetic field and its interactions with the gradient field. “That can actually lead to portions of the magnet vibrating; that leads to heat and that can lead to the magnet losing its superconductivity,” he says. Such a ‘quenching’ event would render the most costly and critical component of the MRI system completely useless.

There are also unanswered questions about safety. To date, imaging experiments in human subjects at up to 11.7 T have given little cause for concern. The primary complaint with ultra-high field is transient disorientation or nausea, and studies suggest that this is largely the consequence of interactions between the gradient fields and the vestibular system in the inner ear, which governs our sense of balance and orientation. But health effects have not been comprehensively studied at 10.5 T or 11.7 T, and regulatory authorities and ethics boards are carefully overseeing human experiments at these facilities. Scheffler predicts that if the Dutch 14-T system successfully goes online, “for the first two or three years, everything will be about safety.”

Only time will tell regarding the scientific returns on this investment. But Petridou is optimistic that even the engineering effort itself will yield substantial dividends, enabling innovations that trickle down to instrument design at lower field strengths. And Shemesh’s experience in working at the frontiers of ultra-high-field MRI have left him bullish about what the future of this technology might hold for neuroscientists. “What you are going to be able to see is going to far exceed what we can see today,” he says. “We’ll be able to address a lot of very, very critical questions about how the brain is wired, connected and functions.”