Resources/MRI fundamentals

MRI physics: from protons to pictures

Every image you plan on the console rests on the same chain of physics — a proton tipped, a signal induced, a frequency decoded into a pixel. This is that chain, built up in the order it happens, for radiographers who want to understand the machine rather than just drive it.

Why hydrogen is the nucleus we image

MRI could, in principle, image several elements. In practice it images hydrogen almost exclusively, and three facts explain why. Hydrogen is abundant — the body is roughly two-thirds water and rich in fat, so hydrogen is the most common atom present, offering an enormous pool of signal. Its nucleus is a single proton with a large magnetic moment, which makes it unusually responsive to a magnetic field and yields a strong, detectable signal. And because water content shifts with oedema, inflammation and tumour, hydrogen is a natural reporter of disease.

A reasonable question is why we do not image electrons, whose magnetic moment is hundreds of times larger. The answer is that in tissue, electrons are almost always paired with opposite spins and cancel out; their resonance would fall in the microwave band that cooks tissue rather than penetrates it; and they relax far too quickly to measure. The proton, by contrast, is a lone unpaired spin that resonates safely in the radio band and holds its signal long enough to read.

Spin, precession and the Larmor frequency

Every hydrogen proton behaves like a tiny bar magnet with an intrinsic property called spin. Placed in the scanner's strong static field — conventionally labelled B₀ and running head-to-foot along the bore — the protons do not simply snap into line. They precess, wobbling around the field like a spinning top leaning in gravity.

The rate of that wobble is the single most important number in MRI, the Larmor frequency: ω₀ = γB₀. The constant γ, the gyromagnetic ratio, is about 42.58 MHz per tesla for hydrogen, so at a common 1.5 T field the protons precess at roughly 64 MHz. That value sets the frequency the scanner must transmit at to have any effect, and it is the hook on which slice selection later hangs.

Net magnetisation: why the signal exists at all

A single proton is undetectable. What the scanner measures is the sum of billions of them in a voxel. Without a field their moments point every which way and cancel to zero. Switch B₀ on and a very slight majority settle into the lower-energy state aligned with the field, producing a small net magnetisation vector — call it M — pointing along B₀.

"Slight" is an understatement: at 1.5 T only about five protons in every million tip the balance. MRI works anyway because a voxel holds so many protons that even that tiny fraction is an enormous number of coherent spins. This is also why field strength matters — a stronger B₀ widens the energy gap, enlarges the majority, and raises the signal-to-noise ratio. Signal is, to good approximation, proportional to B₀.

Resonance: tipping the magnetisation over

M sits along B₀, but a receiver coil can only detect magnetisation that rotates across it, in the transverse plane. So the magnetisation must be tipped away from the main axis. This is the job of a radiofrequency pulse — a small oscillating field, B₁, applied at exactly the Larmor frequency.

The trick is resonance. Because B₁ oscillates in step with the precessing protons, its small torque adds up cycle after cycle, spiralling M down into the transverse plane — a tiny field of microtesla overcoming a static field of tesla. How far M rotates is the flip angle, set by the pulse's strength and duration. A 90° pulse tips M fully into the transverse plane for maximum signal; a 180° pulse inverts it entirely.

Slice selection with gradients

Tipping every proton in the body at once would be useless — we need one slice. A gradient coil makes the field vary linearly along an axis, so the Larmor frequency becomes position-dependent. Transmit an RF pulse tuned to a narrow band of frequencies and only the protons in the matching slab meet the resonance condition and flip. Everyone else stays aligned with B₀ and stays silent. Slice position and thickness are chosen by the centre frequency and bandwidth of the pulse against the gradient strength.

Detecting the signal

Once M rotates in the transverse plane it sweeps past the receiver coil, and a rotating magnetic field through a coil induces a voltage — Faraday's law, the same principle as a generator. That oscillating voltage, at the Larmor frequency, is the raw MRI signal. Every contrast mechanism and every artifact downstream is a modulation of this induced voltage.

Relaxation: where contrast comes from

The moment the RF pulse switches off, the system starts returning to equilibrium by two independent routes. These two relaxation processes are the origin of soft-tissue contrast.

T1 relaxation is longitudinal recovery: magnetisation regrowing along B₀ as protons hand their excess energy to the surrounding molecules. Fat, whose molecules tumble at a rate close to the Larmor frequency, transfers energy efficiently and recovers quickly — a short T1 — so it is bright on T1-weighted images. Free water recovers slowly and stays dark.

T2 relaxation is transverse decay: the spins, initially in phase, gradually lose coherence through their mutual magnetic interactions. Fluid dephases slowly — a long T2 — so it is bright on T2-weighted images, which is why T2 is the sequence that lights up oedema and most pathology.

The operator's controls: TR and TE

T1 and T2 are fixed properties of tissue. What the radiographer controls is how strongly each one shows up, through two timing parameters.

• TR — repetition time, the gap between excitation pulses — sets T1 weighting. A short TR denies tissues full recovery, so their T1 differences dominate. A long TR lets everything recover and erases T1 contrast.
• TE — echo time, the delay before the signal is read — sets T2 weighting. A long TE lets short-T2 tissue fade before readout, exaggerating T2 differences. A short TE reads before those differences appear.

The 180° pulse: inversion and refocusing

Beyond the 90° excitation, the 180° pulse is the other workhorse, used two ways. As an inversion pulse it flips M to point against B₀; as it recovers it passes through zero, and a 90° pulse timed to that zero-crossing nulls a chosen tissue. This is how STIR suppresses fat and FLAIR suppresses CSF to reveal periventricular lesions.

As a refocusingpulse it undoes dephasing. After a 90° pulse the signal decays faster than true T2 because static field imperfections add their own dephasing — the combined rate is called T2*. A 180° pulse reverses each spin's accumulated phase, so the fast and slow spins re-converge and produce a spin echo limited only by true, irreversible T2. Picture runners on a track told to about-face at the same instant: keep their speeds and they all return to the start line together.

Spatial encoding and k-space

Resonance and relaxation give a signal, but not a picture — on their own the scanner would report one number for the whole patient. To build an image the signal has to be located in space, and gradients do it in three steps.

• Slice selection (z) — a gradient during the RF pulse excites one plane, as above.
• Phase encoding (y) — a brief gradient after excitation leaves each row of protons with a different phase offset, encoding vertical position. It must be repeated once per image line, which is why it dominates scan time.
• Frequency encoding (x) — a gradient during readout makes each column precess at a distinct frequency, encoding horizontal position, separated afterwards by frequency analysis.

The encoded signal is stored in k-space, a raw matrix that looks nothing like anatomy — more a galaxy of bright dots. Its centre holds low-spatial-frequency, high-amplitude data that sets image contrast; its edges hold high-spatial-frequency data that sets fine detail and sharp edges. Skip the outer lines and the image blurs; skip the centre and you get edges with no contrast. A Fourier transform — the operation that separates a mixed signal into its component frequencies, like an ear picking single notes out of a chord — converts filled k-space into the grayscale image on the console.

Common pulse sequences

Spin echo is the gold standard for image quality but often too slow for moving organs or restless patients. Three faster families cover most of the rest of clinical practice.

Gradient echo trades image purity for speed: a small flip angle recovers almost instantly, allowing very short TRs, but without a 180° pulse the images are T2*-weighted and sensitive to metal and blood products. Diffusion-weighted imaging is the emergency brain sequence — it images the random motion of water rather than tissue contrast. In an acute infarct the cell pumps fail, water is trapped in swollen cells and its motion is restricted, so it stays bright while normal tissue fades. Confirming a matching dark area on the ADC map separates true restricted diffusion from T2 shine-through.

Image quality: the trade-off triangle

No MRI image is free of compromise. Three quantities compete, and improving one usually costs another, all while scan time has to stay reasonable.

• Signal-to-noise ratio (SNR) — clean versus grainy. Raised by a stronger field, larger voxels, more averages (SNR rises with the square root of NEX, so doubling it means four times the scan), or a narrower bandwidth.
• Contrast-to-noise ratio (CNR) — the actual difference between two tissues. A bright, high-SNR image is still useless if tumour and healthy tissue look identical. Tuned by TR/TE weighting, contrast agents and fat saturation.
• Spatial resolution — sharpness, set by pixel size = field of view ÷ matrix, and by slice thickness. Thinner slices and larger matrices sharpen the image but shrink the voxel and cost SNR.

Artifacts and how to fix them

An artifact is anything in the image that is not real anatomy. Recognising them — and knowing which parameter corrects each — is core radiographer skill, because most are fixable at the console.

MRI safety and bioeffects

MRI uses no ionising radiation, which makes it feel benign — but the fields carry real, occasionally lethal hazards, and the radiographer is the gatekeeper. The risks divide cleanly by which hardware component causes them.

• Static field (B₀) — always on, day and night. Loose ferromagnetic objects become projectiles (the missile effect), and ferromagnetic implants can torque within the body. Screen every person and every item before they enter the controlled zone.
• Gradient fields (dB/dt) — rapid switching induces currents that can cause peripheral nerve stimulation, and the acoustic noise exceeds 110 dB. Slow the sequence if a patient reports painful twitching; hearing protection is mandatory.
• RF field (B₁) — deposits heat, measured as specific absorption rate (SAR) and capped at 4 W/kg whole-body. Enter weight accurately; raise TR or lower flip angle to reduce SAR; pad the patient and never let cables loop or skin touch the bore, to prevent burns.
• Quench — the sudden loss of superconductivity boils off helium that expands 700-fold, risking asphyxiation and frostbite. Reserve the quench button for someone pinned to the magnet or a fire — never for a cardiac arrest, where you evacuate the patient to code them.

The chain, end to end

Read back over it and the machine stops being a black box. A proton precesses at the Larmor frequency; a resonant RF pulse tips the net magnetisation into the transverse plane; the rotating magnetisation induces a signal; T1 and T2 relaxation, shaped by TR and TE, give that signal its contrast; gradients encode position into frequency and phase and fill k-space; a Fourier transform turns k-space into the image. Every parameter on the console is a lever somewhere on that chain — and knowing which lever moves which outcome is the difference between running a protocol and understanding it.