How does a CT scanner work?

A CT scanner looks like a big doughnut with a bed that slides through the hole. The doughnut is called the gantry. Under its smooth cover is a heavy metal disc, well over a metre across, that can spin.

On one side of the disc sits an X-ray tube. A collimator, two blocks of lead, trims its X-rays into a thin fan. Straight across the hole is a curved detector: rows of tiny sensors, often 64 rows of about 900 each, that measure how much X-ray gets through you. (How X-rays are made and why bone stops more of them is the story of XRayClear.)

The whole disc spins around you once every 0.25 to 0.5 seconds. It can keep spinning in one direction because power and data cross over through slip rings: metal bands with brushes rubbing on them, so no cables get wound up. The disc also carries a high-voltage generator and a cooling unit, because almost all the tube's energy turns into heat.

The patient table moves you through the ring in steps of a fraction of a millimetre. In the next room, behind lead glass, a radiographer runs the scan from the console.

One X-ray picture squashes your whole body into a flat shadow: bones in front hide things behind. CT gets around that by taking a shadow from every direction.

As the tube spins, the fan of X-rays passes through the body and lands on the detector. Each detector element measures how much got through along its own straight line. One snapshot gives a projection: a 1D shadow profile of the slice from that angle. A scanner takes about 1,000 of them in every turn.

Stack the profiles one under another, in order of angle, and you get a strange striped picture called a sinogram. Every point in the body traces out a wavy line in it, like a sine wave, which is where the name comes from.

Modern scanners scan in a spiral (also called helical scanning): the table glides through while the tube keeps spinning. The pitch is how far the table moves in one turn, divided by the width of the beam. At pitch 1 the spiral's turns just touch. And with 64 rows of detectors, each turn covers 64 slices at once.

The scanner never sees the slice itself. It only has shadows. How do you get the picture back? This board does it for real, with a famous test slice called the Shepp–Logan head.

Take one shadow and smear it back across the image, along the direction the X-rays travelled. Where the shadow was dark, the smear is bright. Do it for every angle and add the smears up. That is back-projection. The smears pile up where something really is, so the slice appears, but blurry, and with too few angles you see star-shaped streaks.

The fix is to filter each shadow first. The ramp filter makes edges sharper and adds small dips on either side, so the extra blur cancels out when the smears are added. This is filtered back-projection, and for decades it was how nearly every CT image was made. Its neatest version was published in 1971 by G. N. Ramachandran and A. V. Lakshminarayanan in Bangalore, so the filter is called Ram-Lak.

The maths behind all this was worked out in 1917 by Johann Radon, long before anyone could build the machine. Today many scanners use iterative reconstruction, which guesses, checks against the shadows and corrects, over and over.

A CT image isn't really a photo. Every pixel is a number saying how strongly that tiny bit of the body stops X-rays, its attenuation μ. Godfrey Hounsfield turned it into a handy scale, now named after him:

HU = 1000 × (μ − μ_water) ÷ μ_water

So water is 0 and air is −1000. Fat is about −100, soft tissues sit between about +20 and +80, and bone runs from a few hundred to well over +1000. The same tissue gives the same number on any scanner, which is what lets doctors measure things.

But a screen shows only about 256 shades of grey, and our eyes tell apart far fewer. So the viewer picks a window: a level (the middle) and a width. Everything below the window is black and everything above is white. A narrow brain window shows the small difference between blood and brain. A wide bone window shows the skull's fine structure. A lung window shows the air-filled lungs.

Iodine contrast, injected into a vein, stops X-rays strongly, so blood vessels and the heart's chambers light up.

Radiation dose from scans is measured in millisieverts (mSv). We all get some every year from nature: rocks, radon gas in the air, space and even our food. The world average is about 2.4 mSv a year.

A chest X-ray gives about 0.02 to 0.1 mSv. A CT scan takes hundreds of X-ray views, so it gives more: about 2 mSv for the head, about 7 for the chest and about 8 to 10 for the belly and pelvis. Those figures are typical averages for adults; the real dose depends on the scanner, the settings and the person.

Why not just use less? Because X-rays arrive as separate particles, and fewer of them make a noisier, grainier picture. Halve the dose and the noise grows by about 40% (it goes with 1 ÷ √dose). Scanners fight this in clever ways. Tube current modulation turns the tube down where the body is thin and up where it is thick. Iterative reconstruction cleans up noise in the computer, so less dose is needed for the same picture.

CT's great strength is speed. A whole-body scan after a serious accident takes only seconds, and a scan for a suspected stroke can show within minutes whether there is bleeding in the brain. That is why doctors weigh the small radiation risk against what the scan can show.

Dual-energy CT scans with two X-ray energies at once, for example 80 and 140 kilovolts, often from two tubes on the same ring. Low-energy X-rays are stopped much more by heavy atoms like calcium and iodine than high-energy ones are, while water and uric acid barely care. So the ratio of the two readings tells materials apart: a uric acid kidney stone can often be dissolved with medicine, while a calcium stone may need other treatment.

Cardiac CT has to photograph a heart that never stops. The scanner watches the ECG and takes its data in the heart's quietest moment, between beats. Two tubes a quarter turn apart need only a quarter turn to collect a slice, which freezes motion down to about 66–83 milliseconds. With iodine in the blood this becomes CT angiography: a map of the arteries.

Photon-counting CT, first cleared by the US FDA in September 2021, uses a new kind of detector. Instead of turning X-rays into light and measuring the total glow, a crystal of cadmium telluride turns each X-ray straight into a pulse of charge, and the electronics count every photon and sort it by energy. That means sharper images, less noise and energy information in every scan.