CT Physics & Reconstruction
- CT is just an X-ray tube and detectors spinning around you, snapping projections from every angle.
- A computer takes those hundreds of angled X-ray "shadows" and back-calculates a cross-sectional map of how much each spot soaked up the beam.
- That map is reported in Hounsfield units (HU), a scale pinned to water (0 HU) and air (−1000 HU).
- Reconstruction turns raw projections into images; windowing then decides which slice of the gray scale you actually look at.
- Iterative reconstruction lets you keep image quality while turning the dose down — the modern default.
A plain chest X-ray squashes your whole torso into one flat shadow, so the heart, spine, and lungs all stack on top of each other like a badly packed suitcase. CT's entire trick is refusing to do that. Instead of one shadow, it takes hundreds, from every angle, and lets math un-stack them into clean cross-sectional slices. If a regular radiograph is a photo of a loaf of bread, CT hands you the actual slices.
Spin, shoot, repeat
Inside the gantry — the big donut you slide into — an X-ray tube and a curved bank of detectors sit on opposite sides and spin around you together. As they spin, the tube fires and the detectors measure how much beam survived the trip through you from that angle. Dense stuff like bone eats a lot of the beam; air barely touches it. (That eating-of-the-beam is attenuation, and it's the whole signal CT is built on.)
One angle alone is useless — it's just a flat shadow with everything overlapping again. But take a shadow from every direction around the slice and you've boxed the problem in from all sides. Each angle is a clue; stack enough clues and there's only one arrangement of tissue densities that could have cast all those shadows at once.
Modern scanners do this helically: the table glides through the donut while the tube spins, so the beam traces a corkscrew through you rather than stamping out one ring at a time. It's faster and lets the computer reconstruct a slice at any level it likes.
From shadows to a slice: reconstruction
Here's the part that feels like magic but is really just bookkeeping. The scanner collects all those angled measurements (the projection data, or raw data) and has to solve a puzzle: what density goes in each tiny box, or voxel, of the slice so that the math reproduces every shadow we measured?
The classic method is filtered back projection — essentially smearing each angle's measurement back across the image and letting hundreds of overlapping smears reinforce the real structures and cancel out the rest. It's fast and it's been the workhorse for decades.
The newer approach is iterative reconstruction: the computer takes a guess at the image, simulates what shadows that guess would have made, compares them to the real measurements, nudges the guess, and repeats until they match. It's more computational elbow grease, but it cleans up noise beautifully — which is why it's the modern default and a big reason scans can run at lower dose than they used to.
The Hounsfield scale: putting a number on gray
CT's superpower is that every pixel has an actual number, not just a vague shade. That number is the Hounsfield unit (HU), and the scale is anchored to two reference points everyone agreed on:
| Tissue | Approx. HU | Why it's a landmark |
|---|---|---|
| Air | −1000 | The bottom of the scale, by definition. |
| Fat | roughly −100 to −50 | Lighter than water, so negative. |
| Water | 0 | The zero point the whole scale hangs from. |
| Soft tissue / fluid | roughly 0 to ~40 | The crowded middle where most organs live. |
| Acute blood | usually ~50 to 70 | Denser than plain fluid — a key clue on a head CT. |
| Bone / dense calcium | several hundred to >1000 | The dense top end of the scale. |
A measured HU is a real, reproducible number you can put your cursor on. "Is this cyst water or is it something more solid?" is often answered by clicking on it and reading the HU.
Windowing: you can't look at it all at once
A CT slice contains thousands of HU values, but a screen (and your eyeball) can only show a few hundred shades of gray at a time. So we pick a slice of the scale to map onto black-to-white. That's windowing, and it's controlled by two dials:
- Window width = how big a range of HU you spread across the gray scale. Narrow width = high contrast (small differences look dramatic); wide width = everything fits but looks flat.
- Window level = where the center of that range sits — basically, which tissue you've chosen to make mid-gray.
The same raw scan becomes a "lung window," a "soft-tissue window," and a "bone window" — same data, three completely different-looking images. Nothing is recomputed; you're just sliding a viewing window along the HU scale.
Windowing changes the picture, not the patient. If a lesion vanishes when you switch windows, the lesion didn't go anywhere — you've just moved your viewing range off it. Always check a finding in the window built for that tissue.
A common rookie trap: calling something "low density" or "high density" without checking the window. On a punishingly narrow window almost anything can look black or white. Before you trust the shade, glance at the window settings — or better, hover and read the actual HU.
Why this matters at the workstation
Everything you'll later do on CT rides on this foundation. Whether a bleed is acute, whether a nodule is calcified, whether a "mass" is just a fat-density nothing — all of it comes down to reading HU and choosing the right window. The same physics also explains why we sometimes inject iodinated contrast: iodine is dense, so it lights up vessels and enhancing tissue by bumping their HU. And when images look grainy or streaky, the culprits live in image quality and noise and CT artifacts.
If you remember one thing: CT spins, measures attenuation from every angle, and reconstructs a numbered density map — and you decide, with windowing, which part of that map to actually see.