MRI Scanner
Magnetic resonance imaging uses no X-rays. It sets the hydrogen nuclei in the body precessing in step in a strong magnetic field, reads the signal they give off with radio waves, and computes a cross-section from it.
Scroll
A Millimetre of Soft Tissue
A clinical MR image has pixels about 0.5 to 1 mm across and slices 1 to 5 mm thick. Its strength is soft tissue: grey and white matter, muscle and fat, which an X-ray barely tells apart, stand out clearly here.
There is no ionising radiation at any point. The field is 1.5 or 3 tesla, tens of thousands of times the Earth’s; the signal is read with radio waves at about 64 or 128 MHz.
Pixels about 0.5–1 mm · 1.5 T and 3 T · no ionising radiation
Ten in a Million
The body is about 60 % water, and each water molecule has two hydrogen nuclei: protons. A proton has spin, and with it a very weak magnetic moment.
With no field outside, the moments point every which way and add up to nothing. In a field B0 their directions are still almost random, with only a slight statistical lean towards the field: at 3 tesla and body temperature, about ten parts in a million.
A cubic millimetre of water holds about 7 × 10¹⁹ protons, and the tiny lean adds up to a net magnetisation M that can be measured. Each thin arrow here stands for one spin, its lean exaggerated tens of thousands of times; the thick arrow is their sum.
Net polarisation about 5 ppm (1.5 T) · 10 ppm (3 T)
42.58 MHz per Tesla
In a magnetic field, spins precess about the field’s direction at a rate proportional to its strength: 42.58 MHz per tesla. That is about 64 MHz at 1.5 T and 128 MHz at 3 T: the Larmor frequency.
At rest, the spins precess out of step with one another, and the net magnetisation lies along the field. A pulse of radio waves at exactly that frequency tips the net magnetisation over by 90°, into the plane across the field.
Tipped over, it keeps turning at the Larmor frequency, like a spinning magnet, and induces a voltage in a receive coil: the MR signal. The precession here is slowed down hundreds of millions of times.
f = 42.58 MHz/T × B0 · 90° pulse
T1 and T2
The signal does not last. The spins soon fall out of step, and the transverse magnetisation decays with a time constant called T2; meanwhile the longitudinal magnetisation slowly grows back, with a time constant called T1.
Different tissues have different T1 and T2. At 3 T, white matter has T1 about 0.8–1.1 s and T2 about 60–80 ms; grey matter T1 about 1.3–1.8 s and T2 about 90–110 ms; cerebrospinal fluid T1 about 4 s and T2 over 1 s.
Setting the time between pulses (TR) and the moment the signal is read (TE) weights the image towards T1 or T2. On a T1-weighted image the fluid is dark; on a T2-weighted image it is the brightest thing. That difference is contrast.
T1-weighted: TR 500 ms, TE 15 ms · T2-weighted: TR 4000 ms, TE 100 ms
Cut Open: Layer within Layer
The scanner is a set of cylinders on one axis. Outside is the cover; inside it the cryostat, holding the superconducting magnet; further in, the gradient coils, the RF body coil, and last the liner of the bore the patient lies in.
Clinical bores are usually 60 or 70 cm across. The magnet makes a strong, even static field B0; the gradient coils make the field vary with position; the RF coil sends and receives the signal.
Cover · cryostat and magnet · gradient coils · RF coil · liner
Coils at 4 Kelvin
The main magnet is wound from niobium–titanium superconducting wire set in copper. In liquid helium at about 4.2 K (−269 °C) its resistance is zero. An outer set of shielding coils, carrying current the other way, keeps the field close to the magnet.
Once the field is ramped up, the current runs on in the closed superconducting loop with no power supply, so the field is always on, night and day. A vacuum and a cold shield keep heat out, and a cold head on top condenses evaporated helium back into liquid.
If part of the coil stops superconducting, the stored energy turns to heat within seconds to tens of seconds, and the helium boils off: a quench. The gas leaves the building through the quench pipe.
NbTi superconductor · liquid helium at 4.2 K · persistent current
Position Becomes Frequency
There are three gradient coils: x, y and z. Switched on, each adds to B0 a field that changes in step with position, so the Larmor frequency differs from place to place. A gradient of 40 mT/m shifts the frequency by about ±170 kHz 10 cm either side of the centre.
A gradient during the pulse means only one slice matches its frequency: slice selection. During readout, position along one direction is told by frequency, and along the other by a phase step added each time: frequency and phase encoding.
The gradients switch on and off within a millisecond, over and over. Their currents sit in B0 and feel Lorentz forces, so the coils shake in time with the switching and knock loudly, up to about 110–120 dB: patients wear earplugs or ear defenders.
Clinical gradients about 30–80 mT/m · slew rates about 100–200 T/m/s
Sending and Receiving
The RF body coil is a “birdcage”: a copper ring at each end and a dozen or more copper rungs between them. The currents in the rungs rise and fall in turn, making a field B1 across B0 that rotates at the Larmor frequency and tips the net magnetisation.
The returning signal is faint, so it is mostly received by coils close to the body: head coils and body coils, each an array of dozens of small coils.
RF energy warms tissue slightly. The scanner limits its power by the specific absorption rate, SAR: in normal mode, no more than 2 W per kilogram averaged over the whole body.
Birdcage body coil · receive arrays · whole-body SAR ≤ 2 W/kg (normal mode)
k-Space
Each signal read out is one line of k-space: what the image holds at one set of spatial frequencies. Each step of phase encoding fills one more line.
Once it is full, a two-dimensional inverse Fourier transform gives the image. The head here is made of a few ellipses; its k-space is computed exactly from a formula, and the image really is computed from it, line by line.
The lines at the centre set brightness and contrast; the lines at the edges set boundaries and detail. The centre alone gives a blurred image with the right shading; the edges alone leave only outlines.
Image = 2D inverse Fourier transform of k-space
Shielded Room, Safety Zone
The scanner room’s walls, floor and ceiling are lined with copper, the door seals with spring contacts, and the window holds a metal mesh: together a Faraday cage that keeps outside radio out. The 128 MHz of a 3 T scanner lies in the aviation radio band.
The field is always on, and reaches beyond the magnet. Ferromagnetic objects brought near it are pulled hard towards the magnet, so the scanner room is the most tightly controlled zone, and metal, implants and belongings are checked one by one before anyone enters.
The dashed line on the floor is the 0.5 mT boundary, the “5 gauss line”, which pacemakers and similar devices traditionally may not cross until cleared; since 2022 the international standard puts that boundary at 0.9 mT. On real scanners the 5 gauss line lies about 2.5–3.5 m to the sides of the magnet’s centre and 4–6 m along its axis.
Faraday cage · 0.5 mT (5 gauss) line
Taken Apart
The layers slide out along the bore: the cover, the cryostat with the superconducting magnet inside, the gradient coils, the RF body coil, the liner, and the patient table. A clinical magnet is about 1.6 to 1.9 m long and weighs about 3 to 7.5 tonnes with its helium.
Whichever layer the pointer rests on is lifted out and named.
Field Strength
Double the field and the Larmor frequency and the polarisation both double, and the signal grows: in theory a 3 T scanner has about twice the signal-to-noise ratio of a 1.5 T one, and the image is less grainy. The price: tissue T1 grows longer, and the RF energy deposited is about four times as much.
Try thisTurn the field from 1.5 T to 3 T: what happens to the Larmor frequency, the polarisation and the signal-to-noise ratio?
What the model leaves out
This is a model of the principle, not of any maker’s design. The layers’ thicknesses and the number and placing of the coil blocks are drawn so the structure reads clearly; the helium is drawn filling its vessel: conventional magnets hold about 1,500–2,000 litres, newer sealed designs only a few. The head is a Shepp–Logan-style test pattern of ellipses, each filled with one tissue’s T1, T2 and proton density; the image is computed from the spin-echo formula and a 2D inverse Fourier transform, with no chemical shift, flow, coil sensitivity or artefacts. k-space is filled from the centre outwards; real sequences often fill it from one edge to the other. The spins’ lean is exaggerated tens of thousands of times and their precession slowed hundreds of millions of times. T1 between 1.5 and 3 tesla is interpolated linearly. The 0.5 mT line is drawn where it fits; its real distance depends on the scanner and its shielding.
Some common misreadings
MRI uses no X-rays and is not ionising: an RF photon at 128 MHz carries about 5 × 10⁻⁷ eV, while ionising a water molecule takes over ten electronvolts. Protons are not little compass needles that each “flip into line” with the field: a single spin is quantum, and in the field its direction is still almost random, with only a tiny statistical lean; what is measured is the net magnetisation of very many spins, whose motion follows the classical Bloch equations. “Resonance” means the radio frequency matches the Larmor frequency, not that the machine vibrates; the knocking comes from the gradient coils. The field is always on: switching off the lights does not switch off the magnet. “3 T” is a nominal value: some scanners actually run at 2.89 T, about 123 MHz.
Sources
- NIST CODATA: proton gyromagnetic ratio γp/2π = 42.577 MHz/T
- A. D. Elster, Questions and Answers in MRI: net magnetisation (a classical vector)
- A. D. Elster, Questions and Answers in MRI: Zeeman splitting (spins are not only “up” or “down”)
- A. D. Elster, Questions and Answers in MRI: the excess of spins at body temperature is a few per million
- L. G. Hanson (2008): Is quantum mechanics necessary for understanding magnetic resonance? Concepts Magn. Reson. A
- A. D. Elster, Questions and Answers in MRI: superconducting magnets (NbTi, 4.2 K, about 1,500 L of helium)
- A. D. Elster, Questions and Answers in MRI: is the field ever turned off?
- A. D. Elster, Questions and Answers in MRI: quench and quench pipe
- J. Manso Jimeno et al. (2023): review of MRI magnet technology, NMR Biomed. (main and shielding coils, helium)
- Philips: the BlueSeal sealed magnet (about 7 litres of liquid helium)
- A. D. Elster, Questions and Answers in MRI: gradient coils
- A. D. Elster, Questions and Answers in MRI: what’s that noise? (Lorentz forces)
- A. D. Elster, Questions and Answers in MRI: the birdcage coil
- A. D. Elster, Questions and Answers in MRI: SAR limits (IEC 60601-2-33)
- G. J. Stanisz et al. (2005): T1, T2 relaxation and magnetization transfer in tissue at 3T, Magn. Reson. Med.
- J. Z. Bojorquez et al. (2017): What are normal relaxation times of tissues at 3 T? Magn. Reson. Imaging
- J. P. Wansapura et al. (1999): NMR relaxation times in the human brain at 3.0 tesla (T1), J. Magn. Reson. Imaging
- Jiang et al. (2023): T1 and T2 of cerebrospinal fluid at 3 T, Magn. Reson. Med.
- C. de Bazelaire et al. (2004): T1 and T2 at 3 T (subcutaneous fat), Radiology
- A.-M. Oros-Peusquens et al. (2019): proton density of brain tissue, Front. Neurol.
- ACR large phantom testing instructions (spin-echo TR/TE, pixel size and slice thickness)
- A. D. Elster, Questions and Answers in MRI: locations in k-space
- L. A. Shepp and B. F. Logan (1974): The Fourier reconstruction of a head section, IEEE Trans. Nucl. Sci. (the head phantom)
- A. D. Elster, Questions and Answers in MRI: why RF shielding?
- ITU Resolution 413 (WRC-23): the 108–137 MHz aeronautical bands
- A. D. Elster, Questions and Answers in MRI: ACR safety zones
- A. D. Elster, Questions and Answers in MRI: projectiles
- E. Kanal et al. (2013): ACR guidance document on MR safe practices, J. Magn. Reson. Imaging
- College of Physicians and Surgeons of BC: MRI safety standard (IEC moved to 0.9 mT in 2022)
- Siemens Healthineers: MAGNETOM Vida 3T (70 cm bore, weight)
- Siemens Healthineers: MAGNETOM Sola 1.5T (gradients, weight)
- GE HealthCare: SIGNA Artist 1.5T datasheet (63.86 MHz, 5 gauss line, magnet length)
- US FDA: benefits and risks of MRI (non-ionising; projectiles)
- US NIBIB: magnetic resonance imaging (noise up to 120 dB)
- D. W. McRobbie, E. A. Moore, M. J. Graves, M. R. Prince: MRI from Picture to Proton, 3rd ed., Cambridge University Press, 2017

