How does an MRI machine work?

An MRI scanner is a huge magnet with a tunnel through it, called the bore. It is about 70 cm wide. You lie on a table that slides you in.

The magnet is made of coils of niobium-titanium wire. Cooled in liquid helium to about 4 K (−269 °C), the wire becomes a superconductor: current flows round it with no resistance at all, so the magnet stays on day and night without a power supply. The helium sits in a giant vacuum flask, the cryostat, and a cold head on top keeps it from boiling away. A clinical magnet makes 1.5 or 3 tesla, roughly 30,000 to 60,000 times Earth's field.

Inside the magnet sit three gradient coils (x, y and z) that tilt the field slightly so the scanner can tell where signals come from. Inside those is the RF body coil, a radio antenna that sends pulses and listens for echoes. A smaller head coil listens closely to your head.

The whole room is a Faraday cage lined with copper, so that taxi radios and FM stations don't leak in. A quench pipe runs to the roof, ready to vent helium gas safely.

About two thirds of the atoms in your body are hydrogen, mostly in water and fat. The nucleus of a hydrogen atom is a single proton, and a proton behaves like a tiny spinning bar magnet.

Normally these little magnets point every which way, so they cancel out. Put them in a strong field, B₀, and slightly more of them settle pointing along it than against it. Only slightly: at 1.5 T the excess is about 5 in every million. But a millilitre of water holds about 67,000 billion billion protons, so even that tiny excess adds up to a measurable net magnetisation.

The protons don't just line up. Like a spinning top leaning in gravity, each one precesses: it wobbles round the field direction. The wobble rate is the Larmor frequency, and it is set by the field: f = 42.58 MHz per tesla × B₀. That is 63.9 MHz at 1.5 T and 127.7 MHz at 3 T, right among FM radio and aircraft radio.

The scanner sends a short burst of radio waves, an RF pulse, at exactly the Larmor frequency. The spins soak it up, like a swing pushed in time, and their net magnetisation tips over by 90°. Now it spins sideways, and a spinning magnet makes a radio signal in the coil. That is the MR signal.

It doesn't last. Neighbouring spins feel slightly different fields, so they drift out of step and fan out: the signal fades. That is T2 decay. Meanwhile the magnetisation slowly grows back along the field: that is T1 recovery. A second, 180° pulse flips the fan over so the spins catch up with each other and briefly line up again: a spin echo.

Every tissue has its own T1 and T2. Fat recovers fast; the fluid round your brain (CSF) is slow to recover and slow to fade. The operator picks the repetition time TR and the echo time TE. Short TR and short TE make a T1-weighted image: fat bright, fluid dark. Long TR and long TE make a T2-weighted one: fluid bright.

If the field were the same everywhere, every proton would sing at the same frequency and the scanner couldn't tell where anything was. So three gradient coils add a gentle slope to the field: a little stronger on one side, a little weaker on the other.

First a gradient along the bore picks one slice: only protons whose frequency matches the RF pulse get tipped over. Then, while the echo is read, a frequency-encoding gradient makes the left side sing lower and the right side higher, like keys on a piano. A brief phase-encoding gradient, different every time, twists the spins up and down the picture.

Each echo fills one line of a grid called k-space. It doesn't look like a body at all. After 128 or 256 lines, a Fourier transform turns it into the picture. The centre of k-space holds the overall brightness and contrast; the edges hold the fine detail.

Switching the gradients on and off thousands of times a second makes the coils shove against the magnet's field (a Lorentz force), so they flex and bang. That's the loud knocking, often above 100 dB, which is why you get earplugs or headphones.

MRI uses no ionising radiation: no X-rays, nothing that can damage DNA. Its risks come from the magnet and the radio waves, and hospitals manage them with strict rules.

The magnet is always on, even at night. Near the bore the field changes very steeply, and anything made of iron or ordinary steel gets yanked in: the projectile effect. A steel oxygen cylinder can fly across the room. So everyone is screened before going in, and equipment is labelled MR Safe, MR Conditional or MR Unsafe. Some pacemakers, implants and metal fragments can move, heat up or stop working, so doctors check each one first.

The radio pulses deposit a little heat in your body. The scanner watches the SAR (specific absorption rate) and keeps it under limits set by the IEC 60601-2-33 standard: 2 W/kg averaged over the whole body in normal mode, and 4 W/kg in first-level mode, which needs closer watching.

In an emergency the magnet can be quenched: its coils are made to stop superconducting, the stored energy heats the helium, and it boils off as a huge cloud of gas that the quench pipe carries out through the roof.

The same scanner can do far more than take still pictures.

fMRI watches the brain work. Active brain cells need more oxygen, and blood rushes in with more than they use. Blood's haemoglobin is slightly magnetic when it has given up its oxygen, and not when it carries it. So a busy area gets a little brighter, by a few percent, a few seconds later. This is the BOLD signal, discovered by Seiji Ogawa in 1990.

Diffusion MRI measures how far water molecules wander in a few thousandths of a second. In white matter, water moves more easily along nerve fibres than across them, so a computer can trace the brain's wiring: tractography. Diffusion scans also spot a stroke within minutes.

MR angiography makes flowing blood bright, often without any injected dye, to show arteries and aneurysms.

MRI is the best tool for soft tissue: brain, spinal cord, joints, ligaments and many tumours. CT is better and much faster for bone, bleeding after an accident and the lungs, and it takes seconds instead of many minutes.