Skip to content
Imaging Evidence Path

Start here

A practical advanced self-study guide for one learner (Adrian): physician with 20+ years out of clinical practice and senior software/CS/IT experience, moving into medical imaging → quantitative imaging / radiomics → clinically rigorous ML → imaging informatics → medical imaging AI. It assumes you can code; it does not re-teach Python, Git, or basic ML.

The course is built as bounded lessons — each is one evening unit with a small study block at the top telling you how long it takes and what kind of effort it needs. Nothing is gated. Jump to whatever interests you tonight.

Where do I begin tonight?

Open one lesson, not a whole chapter. Each lesson is ~15–40 min and tells you its duration, mode (read / watch / hands-on), energy (low / active) and setup at the top.

How much time do I have?

Time Pick Mode
~20 min Why a CT pixel is not HU read
~20 min The array has no location read
~30 min How texture becomes numbers read + a worked matrix
~30 min What is k-space? read + short video
~40 min Open one real CT hands-on (Python)
~40 min Visual QC in Slicer hands-on (Slicer)
~40 min Tran et al. dissection active reading

Low-energy evenings (read/watch, no setup): lessons marked read or watch above, plus the MRI video lessons. Active evenings (Python / Slicer / data): lessons marked hands-on — open one real CT, Visual QC, the RIDER robustness lab.

The chapters at a glance

Each chapter is an overview page + 6–9 lessons + a reference. The overview states the thread the chapter pulls and lists its lessons.

# Chapter Foundational? Independent entry?
1 CT and DICOM yes — the foundation yes
2 Geometry, segmentation & resampling yes — where the bugs live assumes Ch. 1
3 Quantitative imaging & radiomics core assumes Ch. 1–2
4 Statistical learning & validation core (ISLP companion) yes (read alongside ISLP)
5 Imaging informatics — yes (assumes Ch. 1)
6 Medical imaging AI — yes (assumes Ch. 1–2)
7 Research methodology — yes
8 MRI & multimodality foundations — yes (assumes Ch. 1)
9 Clinical deployment & lifecycle — yes (assumes Ch. 4–5)

Foundational (do first): Chapters 1–4. Enter independently any time: 5, 6, 7, 8, 9. A coherent path, only if you want one: 1 → 2 → 3 → 4, then 5/6/7/8/9 as interest dictates.

The end-to-end pipeline (where each chapter plugs in)

flowchart LR
    A["Acquisition /<br/>reconstruction<br/>Ch.1"] --> B["DICOM series<br/>Ch.1"]
    B --> C["Geometry / SEG<br/>Ch.2"]
    C --> D["Resampling<br/>Ch.2"]
    D --> E["Discretisation +<br/>feature extraction<br/>Ch.3"]
    E --> F["Robustness<br/>Ch.3"]
    F --> G["Leakage-safe<br/>modelling<br/>Ch.4"]
    G --> H["Clinical utility /<br/>appraisal<br/>Ch.7"]
    G --> L["Deployment /<br/>monitoring<br/>Ch.9"]
    I["Informatics / PACS<br/>Ch.5"] -.feeds.-> B
    J["Imaging AI /<br/>seg / synthesis<br/>Ch.6"] -.replaces/augments.-> C
    K["3D Slicer<br/>companion"] -.visualises.-> C
    M["MRI / multimodality<br/>Ch.8"] -.extends.-> J

How does TRACE-CT fit in?

TRACE-CT (../trace-ct) is the practical laboratory: a real project whose DICOM, radiomics and validation work is first-class learning material here. Many lessons end with a TRACE-CT connection — a real script or dataset that makes the concept concrete. Treat it as worked examples and an evidence-generating project, not a gate.

How each lesson is built

Most lessons follow: anchor problem → mental model → concrete example → visual → stop-and-think → tiny application → what to retain. Skip the hands-on parts if the concept is familiar; never skip the what usually goes wrong / stop-and-think parts.

Where to find things