Skip to content
Imaging Evidence Path

Chapter 1 — CT and DICOM

This chapter · ~2 h total across 6 lessons · The foundation everything else stands on · energy: low to active · setup: Python + one RIDER slice for the hands-on lesson

Everything in quantitative imaging and radiomics either runs on, or is silently corrupted by, two models: what a CT pixel actually represents, and how DICOM organises the data that carries it. Get these two right and most later bugs become diagnosable; get them wrong and every downstream number is suspect.

The thread this chapter pulls

A CT slice is not a photograph and not a raw measurement. It is the output of a reconstruction algorithm, stored in a file format (DICOM) whose job is to carry both the pixels and the contract that tells you what the pixels mean and where they sit. This chapter builds those two mental models in order:

  • what the scanner measures vs what it reconstructs, and why a stored integer is not automatically a Hounsfield Unit;
  • why one acquisition produces several different images, and why that matters for any measurement you take later;
  • how a folder of hundreds of DICOM files actually organises itself into studies, series and instances — and which identifiers hold that structure together;
  • how to open one real CT and read its contract in pydicom, checked against what 3D Slicer shows.

You do not need to read the whole DICOM standard. You need a small mental model and a handful of tags — and the habit of reading the contract before trusting the payload.

Lessons

  1. What does a CT scanner actually measure? — projections vs reconstruction; why a pixel is an algorithmic output.
  2. Why a CT pixel is not automatically HU — stored value, slope/intercept, and the first silent quantitative error.
  3. One acquisition, several different images — kernel, thickness, dose, contrast phase; why this is a radiomics variable.
  4. How hundreds of DICOM files become one examination — the Patient → Study → Series → Instance hierarchy through a real folder.
  5. Which identifiers hold the system together — Study/Series/SOP/FoR; object identity vs spatial identity.
  6. Open one real CT and read its contract — a hands-on pass with pydicom, RIDER, and a 3D Slicer cross-check.

When you want the dense tag table, the SOP-class list, the failure-mode catalogue and the “what to retain” summary in one place, go straight to the CT & DICOM reference.

How this chapter connects forward

  • Chapter 2 takes “what the pixels mean” and asks where in the patient they sit — origin, spacing, direction, the slice normal. That is where most serious geometry bugs live.
  • Chapter 3 assumes both: a radiomic feature is a function of (image, mask, geometry), so if any of the three is mis-read here, the feature is silently wrong.

If you only have an hour tonight, lessons 1–2 plus lesson 6 (Tasks 1–2) give you the whole core idea: a CT pixel is a reconstructed, encoded value, and you must apply the contract before you trust the number.