Acquisition of images is the hop from a body to pixels: a modality deposits energy (or listens for it), a detector records the response, a computer writes a DICOM. Everything after — the read, the model, the print — eats that file. It is not a glossary of “what is a modality.” It is not a CT-window / HU article. It is not the QA program.
If you meant imaging vs therapy modalities, DICOM modality codes → modalities definition medical. If you meant CT artifacts / window / HU → artifacts in CT / Hounsfield units / CT window settings. If you meant the imaging-QA program (phantoms / ACR) → medical imaging quality assurance. If you meant what a pixel even is → what is an image. If you meant the 3D field those volumes belong to → 3D medical imaging. If you meant PACS / the DICOM handshake → PACS and DICOM.
PYCAD does not sell a scanner. If the acquired study has to live in a clinic viewer or feed a model, that is the imaging piece.
Four cameras, four physics
You do not “take a photo of the inside.” You pick a physical question and a machine that can ask it.
| Modality | Physics | What you get | What you pay |
|---|---|---|---|
| CT | X-ray attenuation, reconstructed into slices | Bone, lung, contrast vessels, trauma. Seconds to a few minutes | Ionizing dose. Soft tissue loses to MRI |
| MRI | Nuclear magnetic resonance of (mostly) water | Brain, cord, joints, myocardium. No ionizing dose | 30–90 minutes, cost, claustrophobia, metal |
| Ultrasound | High-frequency sound and its echo | A live picture: fetus, heart, a needle | Operator-dependent. Bone and air block it |
| PET | A tracer’s decay — function, not anatomy | Where metabolism is hot. Almost always fused to CT/MRI | Radiation + cost. Spatial detail is poor alone |
The clinician’s question picks the camera. A torn ligament is MRI. A polytrauma is CT. A line placement is ultrasound. “Is this FDG-avid?” is PET. Same patient, four different files.
The file is a DICOM
The machine does not write a JPEG and a sticky note. It writes a DICOM: pixels plus the tags that say who, which protocol, which series, which window. Lose the tags and you have a picture you cannot hang, bill, or train on.
A normal hop: order in the RIS → protocol on the console → acquire → the study lands on the PACS → a viewer opens it. Export-to-USB is the failure mode. The handshake itself is PACS and DICOM, not a second acquisition article.
Quality is decided here
A model cannot invent spatial resolution the gantry did not collect. A radiologist cannot read a motion-smeared MRI as if it were still.
- Motion. The cheapest artifact. Breath-hold, gating, a kid who will not lie still. Repeat is dose (CT) or time (MRI).
- Protocol. Wrong slice thickness, wrong contrast phase, wrong coil. You cannot “window” your way out of a protocol that was not the question.
- Calibration. Water should still be ~0 HU. A phantom that drifted is a service call, not a prompt-engineering problem. The program around that is medical imaging QA.
ALARA is the CT / X-ray / PET constraint: enough dose for the decision, not a pretty volume for a slide. MRI and ultrasound skip the ionizing part and pay somewhere else.
After the file, before the model
If the study is going into a training set or a deployed model, three chores sit after acquisition and are not this URL:
- De-identify. Tags and burned-in pixels. A different page.
- Annotate. A person (or a model a person checks) draws the label the net will copy.
- Augment / normalize. Intensity, orientation, the grid. Preprocessing, not a second “how CT works.”
A market size for “image recognition” is not a reason to change a protocol. Dropped.
FAQ
CT vs MRI — which one?
CT when you need speed, bone, lung, or contrast vessels. MRI when you need soft-tissue contrast and can spend the time. The table above is the whole argument.
Why does everyone mention DICOM?
Because a PNG of a slice is not a study. DICOM is how the pixels stay attached to the patient, the protocol, and the archive.
Does PYCAD acquire images?
No. No gantry, no nuclear pharmacy. Viewer / model when the file already exists. Case studies.