Series 1 · The picture
How an MRI makes a picture
This is a slice through a head, seen from the feet, the way a scanner shows it. Nothing was cut and no X-rays were used. It took a magnet, a radio and about of lying still.
The head is not a real person. It is a model (scanner engineers call a test object a phantom) with the measured properties of real fat, brain and fluid. Every picture on this page is worked out from it the way a scanner works its pictures out, so when you change a setting, the picture changes for the real reason.
- Radio energy going in
- Signal coming back out
- Gradients: the knocking noise
Scroll to follow one scan from start to finish.
Series 2 · The machine
The machine that made it
From outside, a scanner is a short tunnel through a casing about 2 m across. You lie on the table, and it slides you in until the part to be pictured sits at the exact centre.
It looks like one object. It is really four machines, one inside another like the layers of an onion: a magnet, a set of weaker magnets, a radio transmitter and a radio receiver. Each does one job, and a scan is those jobs done in a strict order, a few hundred times over.
The scanner on this page is a model too, drawn to the sizes of a typical 1.5 tesla machine. Its magnet is worked out, not sketched; you will see how.
Series 3 · Inside
Peel it open
Scroll to cut the machine open, one layer at a time, from the outside in.
- You, on the table, in a tube cm wide.
- A listening coil, as close as it can get to the part being pictured. Here it is a cage round the head.
- The radio coil: copper strips in the wall of the tube. It sends the pulses.
- The gradient coils: three sets of copper windings cast into a thick tube of resin. They make the knocking.
- The main windings: the magnet itself. Coils of a wire that has no electrical resistance when it is cold enough.
- Shield windings: two more coils with the current running the other way, to pull the stray field back in.
- Liquid helium at −269 °C, to keep the windings cold.
- A heat shield and a vacuum, one inside the other like a vacuum flask, to keep the warmth of the room out.
- A refrigerator on top, which turns helium that has boiled off back into liquid. It is the steady chirping you hear in the room.
By volume the machine is nearly all magnet. The next stops take the layers one at a time and switch them on.
Series 4 · One voxel
A picture made of numbers
Zoom in and the picture turns into squares. Each square is one number: how much radio signal came from a little box of tissue mm across and mm deep.
Keep going, into one box. It holds about hydrogen atoms, most of them in water (you are roughly 60% water). The centre of a hydrogen atom is a single proton, and every proton is a very weak magnet with a north and a south.
Left alone they point every which way, and all that magnetism adds up to nothing.
Series 5 · The magnet
A current that never stops
The main magnet is a coil of wire with a current going round it, the same as the electromagnet on a scrapyard crane. The difference is the wire: an alloy of niobium and titanium that loses all its electrical resistance a few degrees above absolute zero.
Scroll to start it up. First the tank is filled with liquid helium. Then a power supply is connected and the current is raised, slowly, to several hundred amperes. Then the two ends of the wire are joined to each other, and the power supply is wheeled away.
With no resistance there is nothing to stop the current. It goes on circling for years, and the field fades by less than one part in ten million an hour.
The lines are the field of this model magnet, worked out from its windings. Where they run straight and evenly spaced the field is even, and that is where your head goes. The model is even to about parts in a million across a ball 40 cm wide; a real magnet is trimmed to a few. At amperes these windings would need about km of wire.
Series 6 · In the field
What the field does to you
Now put the protons of that voxel into the magnet. Pick its strength, in tesla:
At T it is about times the Earth's field. Two things happen to the protons.
They do not snap into line like compass needles. Each one wobbles around the direction of the field, like a spinning top leaning in gravity. And slightly more of them lean with the field than against it: about extra in every million. (The drawing shows extra in 144, or you would never see it.)
That tiny surplus is all an MRI has to work with. It is enough only because there are so many protons: the surplus in one voxel is still about .
The wobble rate depends on nothing but the field: million turns a second for each tesla. Here that is MHz, .
In the machine the field runs along the tube. Here it is drawn pointing up.
Series 7 · The radio coil
A field that turns
The next layer in is the radio coil. For its shape, engineers call it a birdcage: sixteen copper strips joined by a ring at each end.
It is fed so that the current is strongest in one strip, then the next, round and round the cage. That makes a small magnetic field pointing across the tube and turning like the hand of a clock, million times a second.
That is the wobble rate of the protons, on purpose. A field that turns in step with them keeps pushing them the same way, turn after turn. Scroll, and the magnetism of the head tips over.
The turning field is tiny: about millionths of a tesla, times weaker than the main field. It works only because it keeps in step.
Drawn tens of millions of times slower than it happens.
Series 8 · The pulse
The right pitch, or nothing
Here is the same tip close up, with the main field drawn pointing up. If the frequency of the radio burst matches the wobble, every turn of the field pushes the protons at the same point in their own turn, like pushing a swing in time. The pushes add up, and in a thousandth of a second the surplus magnetism is tipped over on its side.
Miss the frequency and the pushes cancel. Try it:
Being off by 3 kHz is an error of one part in 20,000, and almost nothing happens.
This is the resonance in magnetic resonance imaging. It is also why the scanner sits in a copper-lined room: a radio transmitter on the same frequency would drown you out.
Series 9 · The signal
The body answers
Tipped on its side, the magnetism keeps wobbling, and a magnet swinging past a loop of wire makes a voltage in it. A bicycle dynamo works the same way.
For a moment you are a radio transmitter. The loops of the listening coil light up in turn as the magnetism sweeps past each one. They sit as close to the head as they can, because the signal is only a few millionths of a volt and falls away quickly with distance.
It is a pure tone at MHz that fades within a few hundredths of a second. Everything an MRI knows comes from this fading tone.
Series 10 · T1 and T2
Two ways the ring ends
After the pulse two separate things happen, at different speeds.
The protons fall out of step. Each feels its neighbours' tiny fields, wobbles a touch faster or slower, and the tone fades. The time this takes is called T2.
The magnetism stands back up along the main field, as the protons hand their energy to their surroundings. That takes longer, and is called T1.
The numbers differ because water behaves differently in each tissue: tumbling freely in fluid, hemmed in among fats and proteins. That is the whole trick. An X-ray sees how dense you are. An MRI sees how the water in you is living.
Each dot on the charts is drawn as bright as that tissue would be at that moment.
Series 11 · The echo
The echo
Most of the early fading is not T2 at all. No magnet is perfectly even, so protons in different spots wobble at slightly different rates and fan out within a few milliseconds. That part can be undone.
Picture runners leaving a start line at different speeds. Shout "turn round!" and the fastest, who has gone furthest, has furthest to come back. They all reach the line together.
A second radio pulse, twice as big, is the shout. Wait as long again and the protons are back in step and the signal returns: a spin echo, found by Erwin Hahn in 1950. The time from the first pulse to the echo is called TE.
Whatever is still missing at the echo was lost to true T2, and that is the part that tells tissues apart.
Series 12 · The gradient coils
Three weaker magnets
So far the signal says what the tissue is like, but not where it is. The next layer of the machine is there to answer "where".
Cast in resin just inside the big magnet are three more coils, one for each direction. Each is wound so that it adds a little to the main field on one side of you and takes a little away on the other, with no change at the centre. The field becomes a ramp.
The ramp is gentle: thousandths of a tesla per metre here, so at the wall of the tube the field is only parts in a thousand away from its value at the centre. But it has to be switched on and off in a fraction of a thousandth of a second, and that takes hundreds of amperes.
A wire carrying a current inside a magnetic field is pushed sideways. That is how a loudspeaker works, and this is a loudspeaker weighing most of a tonne inside a very strong magnet. Every switch of the current jolts the tube. That is the knocking.
The flexing is drawn many thousands of times too big, and the ramp several hundred times too steep.
Series 13 · Where
Pitch is place
In 1973 Paul Lauterbur saw what a ramp like that is good for.
Wobble rate follows field strength, so on a ramp every position sings at its own pitch: kHz higher at one edge of this picture than in the middle, and as much lower at the other.
The coil hears all those pitches at once, as one echo. A piece of mathematics called the Fourier transform sorts a sound into its pitches, as your ear sorts a chord into notes. Sorted, the echo becomes a profile: how much signal came from each strip across the head.
Series 14 · The slice
Choosing a slice
The same trick picks the slice. Turn on a gradient from head to toe during the radio pulse, and only one thin sheet of you is at the right pitch. The rest ignores the pulse; you saw how picky resonance is.
Change the radio frequency and a different sheet answers. A slice mm thick is a band of pitches only kHz wide. Nothing has to move: for a head, thirty or so slices are collected one after another.
Series 15 · One knock
One knock, in slow motion
Now run all the layers together. One knock of a scan is the same five steps every time. Scroll through them. The chart under the machine is how physicists write a scan down: time runs left to right, with one line for each coil.
- Pick the slice. The head-to-toe gradient comes on and the radio coil fires. One thin sheet of the head is tipped.
- Mark the rows. A short kick from the front-to-back gradient, a different size on every knock.
- Turn round. The second, bigger radio pulse, which will bring the echo.
- Listen. The left-to-right gradient comes on, so that pitch means place, and the listening coil records the echo. That is one row of raw data.
- Wait for the magnetism to stand back up.
The first four steps take about thousandths of a second (ms). The wait is the rest of the ms: this scanner spends more than nine-tenths of its time waiting. A real one fills the gap by doing the same to other slices.
Series 16 · The raw data
One knock, one row
One echo gives one profile, not a picture. To tell top from bottom, the scanner repeats the knock again and again, each time with a row kick of a different size.
Each echo is written into one row of a grid of raw data. Physicists call the grid k-space.
Scroll to fill it. The picture on the right is the Fourier transform of whatever has been collected so far. It stays almost empty for a long time, then arrives nearly all at once as the middle rows come in.
The wait between knocks is called TR. This scan repeats every ms and needs rows: .
Series 17 · The raw data
What the grid holds
No point in the raw data is a point in the head. Every point holds a little of the whole picture, like one note of a chord. Throw parts away and see:
Scans are made in rows, minutes apart from first to last, so a scan cannot be paused and anything that moves leaves its mark everywhere.
Series 18 · Contrast
Same head, different questions
The operator chooses two waiting times, and they decide what the picture shows.
TR, the wait between knocks. Keep it short and only tissues with a short T1 have stood back up in time, so they are bright.
TE, the wait for the echo. Make it long and only tissues with a long T2 are still ringing.
There is a small patch of swollen tissue in the model, marked by the corner ticks (on the right of the picture, which is the patient's left). Swelling means extra water. It is invisible on a T1-weighted picture and bright on a T2-weighted one. FLAIR adds one more pulse, timed to catch fluid at the instant it has no signal, so a bright patch stands out even beside a fluid space.
No single picture shows everything. That is why a scan is a list of series, and why it takes half an hour.
Series 19 · Trade-offs
Sharp, clean, fast: pick two
Smaller voxels show finer detail, but each holds fewer protons, so the picture is grainier, and more rows take longer. Repeating the scan and averaging beats the grain down, but four times the scanning only halves it.
A stronger magnet helps: twice the field gives roughly twice the signal. That is what a hospital is paying for with a 3 tesla scanner.
Series 20 · Safety
Because the current never stops, neither does the field: nights, weekends and power cuts included. Steel is pulled harder the closer it gets, and then it flies. Scissors, oxygen cylinders, chairs and floor polishers have all ended up in the bore.
Hence the questions and the metal check before you go in, and the line on the floor where the field has dropped to 0.5 millitesla, ten times the Earth's. The dashed line is that line, worked out for this model magnet. See what the shield windings are for:
Implants and equipment are labelled:
What an MRI does not use is X-rays, or any radiation able to damage DNA. Its radio waves warm you slightly, and the scanner limits how much. The noise is commonly 80 to 110 decibels, so you are given earplugs. Some scans add an injection containing gadolinium, a metal that shortens T1 wherever the blood carries it, so those places turn bright.