Medical imaging starter is the on-ramp for a person, not a modality textbook. Computer vision on one side, medical files and just-enough anatomy on the other. Most people arrive with only one. This page is which hole to fill first, and which live URLs to open. It is not “what is a CT.” It is not “what is a pixel.”
If you meant body → pixels, the four cameras → acquisition of images. If you meant the 3D field (volume, not a product) → 3D medical imaging. If you meant imaging vs therapy, DICOM modality codes → modalities definition medical. If you meant what a pixel / RGB / monochrome file is → what is an image. If you meant the DICOM file and the protocol → what is DICOM.
PYCAD builds custom web DICOM viewers and imaging models. It is not a bootcamp and it does not sell a “starter” product. The old free-ebook landing (/medical-imaging-ebook/) 404s. The old gated notebooks zip (/30-30-medical-imaging-notebooks/) 404s; the list that remains is 30 days, 30 minutes.
Two holes, one job
A working imaging engineer has to open the file and know what the organ is supposed to look like on that camera. A clinician who can read the study still has to know what a convolution is before they can train a model that does not leak the patient ID. You do not need a second degree. You need the missing half, on the project in front of you.
| You already are | Learn first | Open these, in this order |
|---|---|---|
| Computer vision / ML | DICOM / NIfTI as the file. One organ’s anatomy, on the modality that study uses. Window/level is not Instagram | What is DICOM → acquisition → what is an image (pixels) → one anatomy post for the organ you were hired on |
| Clinician / biomedical | Supervised learning, then a CNN, then one small project on public PNGs before you touch a DICOM | Andrew Ng’s ML specialization and the CNN course on Coursera (named, not ranked) → CNNs explained → CNNs or ViT (a real 2023 run, caveated) |
| Neither | Both columns. File formats before models. One organ, not all of Gray’s | The CV row, then the clinician row. Do not start on a “foundation model for every scan” |
If you already write models
Learn the envelope before you learn a new backbone. A JPEG chest network that ignores the DICOM window will not survive a real series. Anatomy is project-scoped: hand bones if the job is a hand; liver segments if the job is a liver. You do not need to become a radiologist to segment a mandible.
- Files. What is DICOM. NIfTI shows up the moment you leave the hospital viewer; conversion is a later cluster, not this page.
- How the camera made the pixels. Acquisition of images. Modalities (codes + therapy) is a different job — modalities definition medical.
- A reading list, not a zip. 30 days, 30 minutes still names five notebooks. amine0110/pycad
tutorials/is the public sibling. - Video. youtube.com/c/pycad (DICOM / NIfTI courses). DigitalSreeni is a long public channel of imaging projects; it is not a PYCAD product.
If you already read studies
Start outside medicine. Classify cats. Detect a box. Segment a leaf. Then take the same code to a public medical set. The file format is the new part, not the loss.
- Machine Learning Specialization (Coursera / DeepLearning.AI).
- Neural Networks and Deep Learning and Convolutional Neural Networks.
- Then a medical repo that still opens: amine0110/CNNs_vs_ViT (the write-up is this post), liver segmentation (MONAI), the mandible write-up on mandible segmentation.
Public sets for that first medical project live on medical image datasets. Do not start on a hospital PACS export.
What this page is not
- Not 265. Acquisition (the physics hop) stays on that URL.
- Not 247. The 3D field stays on that URL.
- Not 527. Imaging vs therapy / DICOM modality codes stay on modalities definition medical.
- Not 612. Pixels / RGB / monochrome stay on what is an image.
- Not a course, an ebook, or a zip. Those landings 404. The newsletter at newsletter.pycad.co is a mailing list, not a replacement zip.
If the missing piece later is a viewer or a model, that is the imaging piece. Case studies.