The most powerful MRI scanner in the world delivers its first images

T1 is the time it takes for the energy that B1 adds to the nuclear spins to leak back out of the nuclei. It generally happens faster when the energy per nucleus, which is proportional to the degree to which B0 has split the energy levels, is low. If I had to guess, it's probably because there are more weak couplings by which a nucleus can get rid of a little energy than there are strong couplings that can support a big jump. One of the major things that limits the speed at which NMR can be done is the fact that you have to wait for this energy to leak out before you can put more in for another round of imaging. (Maybe there is a way around this but if so they never taught me about it.)

T2 is the time it takes for the excited spins to fall out of phase with each other. It happens faster when there is a greater diversity of electronic influences on each nuclear environment - a bunch of identical atoms in a perfect lattice would never go out of phase. If you "zoom in" on T2 and study it with incredible precision you can pick apart each distinct environment around each population of similar nuclear neighborhoods. A population of identical neighborhoods, in practice, means a chemical. Chemists use that a lot but it's not done in imaging.

This explanation is extremely simplified, and the details here depend on the situation. Relaxation is a critical part of MRI and NMR, and there are different mechanisms that contribute to it.

The sensitivity of magnetic resonance methods increases with the magnetic field strength to the power of 3/2. But other effects like the mentioned relaxation cause certain practical issues at higher fields. So you don't always get the full benefit of the higher magnetic field, depending on what exactly you are doing.

Now, for B0. The simple explanation is this: as alluded to above, we're all made of very different stuff. Bone, air, fat and muscle all have very different electrodynamic properties. At their interfaces, there's a "rule" that comes from Maxwell's equations† that means that you _must_ have a discontinuity in both the normal component of the electric (D) field in media and the tangential component of the magnetic field (H). This isn't negotiable, under the assumption that your patient does not have a net overall charge or current flowing through them. What this means, practically, is you get tiny bound surface charges or currents appearing at the interface of different materials. These then change the magnetic field homogeneity -- how B_0 varies over space -- because they create magnetic fields that add to it (the principle of superposition). Remember, all of MRI is based on having extremely homogeneous fields so that you can make a mapping between frequency and phase of your signal to the space of your sample. Inhomogeneities destroy that -- they can e.g. make two places look the same if they should be different, for example. B0 homogeneity is a big problem: we need the field to be uniform across the sample.

Practically, what that means is twofold. One, for big changes -- caused by air, or a dental filling typically -- you get large, far-field effects that distort the image macroscopically. The maxillary and ethymoid sinuses (in the front of your face) together with the hard palette make it harder to image certain regions of the brain -- the bits that are basically directly above the roof of your mouth. This is _entirely_ due to the difference in the dielectric properties of air and "meat". Lung imaging with MRI is very, very challenging (but possible!). What works well for the brain does _not_ work at all, really, on the long -- and especially not at ultrahigh field. Even something as "simple" as cardiac imaging becomes more difficult, because the heart contains oxygenated and deoxygenated blood in close proximity -- and they have different dielectric properties.

Secondly, all tissue on a microscopic level is inhomogeneous. Cells have bilayers and a complex microstructure. Organs have microstructures made up of different types of cells. The liver has a a beautiful microstructure, and a lot of iron in it. This causes a decrease in the effective static field inhomogeneity -- which manifests itself in terms of an NMR relaxation parameter called T2* -- meaning, in short words, that the MRI signal vanishes more quickly over time than it does at lower fields. That's why livers tend to look a bit darker than their surrounding organs on some types of MRI scans. If you play a single RF pulse and acquire signal on an NMR tube of pure water in a chemistry lab, you'll get a voltage you could measure with a hand-held AC voltmeter appearing for seconds after the initial excitation. Put a sphere of water in a 12T MRI scanner, and frankly you're doing well to have signal there after ~500 ms. In vivo, that's likely to be ~30-50 ms. That makes actually doing the experiments hard - your [gradient-echo] signal vanishes quickly.

The whole B0 homogeneity thing is a coupled problem, best solved by considering Laplace's equation for the scalar magnetic potential (\nabla^2 \psi = 0). This famously has a solution space that can be expressed as an infinite sum of spherical harmonic functions. So, MRI scanners (yes, even clinical ones!) have a series of loops of wire that generate B-fields that approximate the first few spherical harmonics -- in Cartesian coordinates, they're weird things like x^2-y^2 and look like the pictures of the orbitals for electrons inside a hydrogen atom, if you've ever seen that. Patient goes in scanner, B0 field inhomogeneity is measured, magic regression‡ works out the right coefficients to dial a current into each coil, and the field is magically more homogeneous.

Now, the trouble with this is that the spherical harmonics are a sum until infinity. And you can't build an infinite number of coils. The first few components get you most of the way there...but not all. And again, at higher field building effective shim coils, working out what your artefact is and where it comes from, and so on, all become more difficult.

</end lecture>

MRI is complicated -- it spans mathematics, physics, chemistry, engineering, biochemistry, and medicine. Frankly, that's why I like working in it as an academic area -- there's always something to learn. There are about 6000 others who do so technically, and we have a big, week-long worldwide conference once a year to both teach this stuff (where I have lectured) in quantitative detail, and show off our latest and greatest results (the usual big "evil" companies typically take academic work and put it in their products -- which I find rewarding, as it means your work can have a difference quickly and you don't have to deal with the FDA). I'm sure CEA / ISEULT will present there and I am genuinely excited that their mega-machine has finally been built. But there are a lot of challenges to overcome. Of course, there are also a huge number of opportunities -- it is (hopefully!) going to make it possible to image at higher spatial resolution, and also be much better for MR spectroscopy, and reveal information about low concentrations of biomolecules. But I guess the main point of my comment originally last night was trying to convey "the journey is really just beginning" and not "expect this in your local hospital in five years time".

---

[1] https://sci-hub.st/10.1002/cmr.b.21319

† Specifically Gauss's law and Ampere's law at the boundaries for the D or H fields giving rise to the rule that there must be a discontinuity in D, with the difference normal components being equal to the bound surface charge density; and the difference in the tangential part of the H field being equal to the bound surface current. Putting these together with the constitutive relationships for D and H, and you get that the surface current is equal to B_{tangential, 1} / µ_1 - B_{tangential, 2} / µ_2 = J_s. In biology, µ is approximately fixed...so the surface current changes.

‡ Or a patient graduate student / radiographer...

B1 is the oscillating RF magnetic field that is responsible for both delivering power into the sample / patient, and once their nuclei have received that and undergone the magnetic resonance phenomenon, they emit RF at the same frequency. We / the MRI community divide this field into two: B1+ and B1-, for transmit and receive (and they also happen to be left and right-handed circularly polarised fields). It's worth pointing out that a physical device -- a lump of copper with capacitors in it -- called an RF coil transmits B1+ and receives B1-. Because we want B1+ to be homogeneous and uniform and B1- to be sensitive, in the last ~20 years, most clinical scanners now have a separate "body coil" for transmission and multiplee anatomically shaped ones ("breast/thoracic/spine/knee/ankle/wrist/whatever-coils") for reception. Multiple because you can noise-optimise the size of each little loop, and you can also do tricks ("parallel imaging" or "compressed sensing") to exploit their different spatial locations and speed up the scan.

Now, at low field, say at 1.5 T (a typical clinical MRI scanner's field), the Larmor frequency for protons that are most often of interest to the operator of the scanner is objectively low (about 60 MHz for 1.5T). By 'objective low', I mean "its wavelength is a fair bit longer than the instrument". Mathematically, if you took a course that involved Maxwell's equations, we're in the quasi-static regime: \nabla \times B = µ0 ( J + \epsilon_0 ∂E/∂t) and at low field ∂E/∂t ≈ 0 compared to J. So, Biot-Savart's law works for describing the coils – you put a current in them (and quite a large one -- 1-3 kV voltages are not unusual for transmit inside a body coil on a device that has a minimally low resistance) and get a magnetic field out. Everything is DC, doesn't interact, and that magnetic field is what permeates through the patient. The optional difference in electronics aside, there's no substantial difference between B1+ and B1- and all is hunky-dory and comparatively well understood.

Now, at high fields, a bunch of stuff happens. Firstly, the frequency goes up -- at ~12 T, it is ~500 MHz -- and with that, the wavelength goes down. At 12T (or 9.4T, or 7T, the other commonly-used high-field strengths where this is relevant), it's well below the size of a person, and rather than having a quasi-static problem, you now have a full-wave electrodynamic problem in which you can't just ignore the displacement current, ∂E/∂t ≠ 0. As well as making it harder to design RF coils, this actually means that you cannot ignore the fact that it's a wave -- to a first approximation you'll get nodes and anti-nodes in the tissue on a human length-scale. Not only that, but all of the really important electrodynamic parameters -- the µr's and \epsilon_r's that determine what speed the wave propagates at, and so on -- are (a) actually complex numbers; (b) different in different tissue types / diseases; and (c) not as well measured at high field as they are at low field. Another effect emerges from this at ultrahigh field: B1 twist. B1+ starts being ≠ B1-* (where the * denotes complex conjugation) and you get some rather "interesting" patterns appear. [1] From an electrical engineering point of you, you've got a near-field antenna design problem into a fundamentally unknown electrodynamic environment (unknown, because to some extent you are measuring it! Is that a cystic, fluid-filled region in the brain, or a dense, cellular tumour?)

So, at ultrahigh field, the only thing you can do is change your approach completely. Rather than having one homogeneous, optimised transmission device, you have multiple -- transmitting in parallel ('pTx') -- and change the phases and amplitudes between them to optimise the homogeneity in the response. This has the downside that you have to be careful not to accidentally start to heat your patient, and so throw in a lot of work, simulations and measurements about safety here. Rather than looking like a loop of wire, coils now tend to look like e.g. fractionated dipole antennas -- which makes them very sensitive to their electrodynamic tuning and exactly the value of µr and \epsilon_r (or equivalently the conductivity) of the sample. Fat is different here to muscle, and that difference is bigger at ultrahigh field...so you now see people selling "dielectric pads", foam-wrapped pads of a material with a favourable \epsilon_r to literally pad around the patient if you have a problem. These too have their own safety concerns, so you end up spending a week running a full-wave FEM simulation of Maxwell's equations prior to destructive heating testing before doing it on a human. The whole thing becomes a bit of a pain.

[1/2 -- I never knew HN had a comment character limit until now!]

> Medical physicist. Nerd. [...]

But they are. CT uses pretty high doses, and each CT scan increases lifetime risk of cancer by order of magnitude 0.1%. CT scans absolutely should not be done willy-nilly.

https://www.clovisopenmri.com/blog/512 has some nice examples, too.

Does it? Couldn't you let go of that assumption with heaps of compute power?

The radiology report said, “No significant incidental findings identified,” so that’s good at least.

Might be safe anyway

Moving in the scanner doesn’t cause you much sensation at all.

However move quickly near the entrance to the bore and you get vertigo.