How to Annotate Medical Imaging Figures: Labeling Lesions on CT, MRI, and Chest X-ray

The first decision is whether to mark up the actual DICOM-derived image or to draw a clean schematic beside it. They answer different questions.

• Annotate the real image when the point is the image — the texture of a ground-glass opacity, the exact margin of a mass, the true appearance a trainee must learn to recognize. Radiographic evidence loses its force the moment it stops looking like a real scan.
• Draw a schematic when you are teaching a concept rather than a specific case — the typical location of a watershed infarct, the zones of a lung, how a herniation displaces midline structures. A schematic is honest precisely because no one mistakes it for a patient's actual scan.
• Use both, side by side for the strongest teaching figure: the real scan with minimal annotation on the left, a labeled schematic keyed to it on the right. The reader sees the finding and the idealized model in one glance.

A common mistake is drawing a schematic and styling it to look photographic, or painting a lesion onto a real scan. Both blur the line between what was observed and what was illustrated. Keep that line visible.

Labeling lesions on CT, MRI, and chest X-ray

The imaging modality changes what your annotation has to respect.

CT. Findings depend on window settings, so state the window (lung, mediastinal, bone) in the caption — an annotation is meaningless if the reader cannot reproduce the contrast you saw. Use arrows for focal lesions and dashed ROI boxes when you cite Hounsfield units. Keep annotations off the finding itself; point from the margin.

MRI. Always label the sequence (T1, T2, FLAIR, DWI) directly on or beside each panel, because the same lesion looks opposite on different sequences. When you show a multi-sequence panel, use the same arrow position across panels so the reader tracks one lesion through its changing signal.

Chest X-ray. Radiographs are low-contrast and busy, so annotation discipline matters most here. A single well-placed arrow beats a scatter of marks. Mark laterality (a persistent "R"/"L" marker), and if you crop, keep enough of the ribs and mediastinum for the reader to orient.

A reusable annotation layout

Whatever the modality, a clean figure tends to share the same anatomy:

1. The image, oriented conventionally. Keep radiological orientation (patient's right on the viewer's left). Do not flip a scan to make a layout tidier — laterality errors are dangerous.
2. One primary callout per finding. An arrow or arrowhead pointing from clear background into the lesion margin, never covering it.
3. Short labels, keyed to the caption. Use letters or brief tags on the image and spell them out in the caption, rather than crowding the image with sentences.
4. A scale reference and orientation. A scale bar or the scanner's own measurement, plus sequence/window/laterality metadata.
5. Consistent color. Pick one high-contrast annotation color and use it for all findings of the same type; reserve a second color only for a genuine second category.

An example prompt for a schematic

When you need a schematic companion rather than a marked-up scan, describing it in plain language gets you a first draft fast. A prompt like this gives an illustration tool enough structure to work with:

A schematic companion to a CT scan — a labeled diagram, clearly not a photograph.

Then edit the labels and geometry to match the concept you are teaching. The tool gets you past the blank canvas; the clinical judgment stays yours.

The ethics of editing medical images

This is where a good figure and a misleading one diverge. Journals and radiology societies draw a hard line between display adjustments and manipulation, and your figure has to stay on the right side of it.

• Allowed, if applied to the whole image and disclosed: cropping, adjusting brightness/contrast or window/level, and standard reconstructions. These change how the image is displayed, not what it contains.
• Not allowed: erasing or adding anatomy, painting in or brushing out a lesion, cloning regions, or selectively adjusting one part of an image to emphasize a finding. That fabricates evidence.
• Always required: remove patient identifiers (name, MRN, dates burned into the pixels, facial features on 3D reconstructions) before the image leaves the clinical system, and confirm you have the consent and IRB/ethics clearance your journal requires.
• Keep annotations on a separate layer from the pixel data, so it is always obvious what is the scan and what is your markup — and so a reviewer can see the unannotated original if asked.

The test is simple: could you show the annotated figure and the raw scan side by side without embarrassment? If yes, your edits are display, not deception.

A final checklist

Before the figure goes into a slide deck or a manuscript, confirm:

• Is every finding you discuss marked, and is nothing marked that you do not discuss?
• Are modality, sequence/window, and laterality stated so the reader can reproduce your view?
• Is it obvious which panels are real scans and which are schematics?
• Have all patient identifiers been removed, and is consent/ethics in order?
• Would the raw, unannotated image support every claim your markup makes?

For related layout and revision advice, see How to Draw Scientific Figures: Complete Guide, the AI prompt guide for writing better generation prompts, and Fixing Figures After Peer Review for handling reviewer requests on imaging panels.

When you need a schematic to sit beside a real scan, or a clean anatomical illustration for a case report, start a draft with the medical illustration generator and refine it until every label is true. A medical figure earns trust the same way a diagnosis does — by pointing at exactly what is there, and nothing that is not.