3D Printing Medical and Dental Models from CT and MRI Scans: DICOM Segmentation Workflow
A growing number of makers with resin or FDM printers are turning patient CT and MRI scans into physical anatomical models — skull replicas for pre-surgical visualization aids made for a surgeon by a hobbyist relative, dental study models printed from intraoral scanner exports, or simply a 3D-printed vertebra for a biology class. The imaging pipeline looks intimidating at first because it crosses from medicine into CAD, but the actual software (3D Slicer) is free, open source, and not meaningfully harder to learn than a slicer. This guide walks through turning a stack of DICOM images into a clean, printable STL, and is equally useful whether you're printing a dental arch model from an intraoral scan or a reference skull from a CT export.
What You're Actually Allowed to Print
Before touching any software, draw a hard line in your head between two very different categories of output, because the segmentation workflow is identical for both but the stakes are not:
- Reference and educational models — anatomical visualization aids, teaching models, pre-surgical "rehearsal" copies a surgeon uses to plan an approach, patient communication aids. This is the overwhelming majority of hobbyist medical 3D printing and it's what this guide covers.
- Patient-contact or implantable devices — surgical guides that touch tissue, cutting jigs used intraoperatively, splints, prosthetics, or anything implanted. These are regulated medical devices in most countries (FDA Class I/II clearance pathways in the US, similar frameworks elsewhere) and require biocompatible, traceable materials, a quality system, and in most cases clinical sign-off. Printing one of these on a desktop printer for actual clinical use without that regulatory path is not a gray area — don't do it.
If a clinician asks you to print something for actual intraoperative use, the right answer is to point them at a cleared medical 3D printing service, not to print it yourself from a home file.
Getting the Scan Data
Patients in the US and most other countries are legally entitled to a copy of their own imaging. Hospital radiology departments and most imaging centers will provide a CD or a download link containing the full DICOM series on request — ask for "the DICOM files," not just the PDF report. Dental impressions are simpler: most intraoral scanners (iTero, 3Shape, Medit) export directly to STL or PLY, skipping the segmentation step covered below entirely, so if you're starting from a dental scan you can jump straight to the mesh cleanup section.
A DICOM series isn't one file — it's a folder of hundreds of 2D slice images plus metadata. Each slice carries the patient's name, birth date, and medical record number in its header. If you're going to share the files, post screenshots anywhere, or even just store them outside a secure folder, strip that metadata first. 3D Slicer's DICOM module can anonymize on import, and standalone tools like dcm2niix or the free DicomCleaner utility will strip identifying tags from a whole folder in one pass.
Segmentation in 3D Slicer
3D Slicer (slicer.org) is the de facto standard free tool for this, used in actual hospital research labs as well as by hobbyists. The core workflow:
- Load the series. Drag the DICOM folder onto Slicer, or use the DICOM module's import button. Slicer will group slices into a proper 3D volume automatically.
- Open Segment Editor. This is where you tell Slicer which voxels belong to the structure you want to print.
- Threshold first. Bone on a CT scan sits in a predictable Hounsfield unit range (roughly 300–3000 HU depending on density); the Threshold effect lets you paint in everything inside that range in one click. This gets you 80% of the way for CT bone work. MRI has no equivalent fixed scale, so soft-tissue segmentation from MRI relies more on manual painting and the "Grow from seeds" region-growing tool.
- Clean up by hand. Use the Scissors, Islands, and Paint effects to remove floating noise, disconnected specks, and anatomy you don't need. This is the slow part — budget real time here, especially for anything with soft tissue.
- Smooth. Slicer's Smoothing effect (Gaussian or Median, modest kernel size) removes the stair-stepping that comes from slice thickness without erasing real anatomical detail. Over-smoothing is a real risk on thin structures like turbinates or small vertebral processes — check before and after on a cross-section view.
- Export to model. "Segmentations → Export to Files" or the Export to Model button generates an STL directly from the segmentation.
From Segmentation to Printable Mesh
Medical segmentation STLs are almost never watertight on the first export — small internal voids, self-intersecting triangles, and non-manifold edges are the norm, not the exception, because the segmentation is built from a voxel grid rather than drawn as a solid. Run the export through a dedicated mesh repair pass before it ever sees a slicer:
- Blender's 3D-Print Toolbox add-on (built in, just enable it) flags non-manifold edges, intersecting faces, and thin walls in one analysis pass.
- Meshmixer or Autodesk's (free) Meshmixer successor workflows handle automatic hole-filling and remeshing well for organic anatomical shapes.
- Decimate aggressively. A raw segmentation export can carry millions of triangles for a single bone. Decimate to a few hundred thousand triangles before you do anything else — your slicer will thank you, and visually the loss is usually invisible for anatomical reference prints.
Scale and orientation matter more here than in most printing: verify the model's real-world dimensions against the original scan's reported measurements before you print (Slicer reports true physical units throughout, so an accidental unit mismatch during STL export, like mm vs. cm, is the most common way a skull ends up printing at the size of a grapefruit or a beach ball).
Printer and Material Choice
Use caseBest processWhy Full skull / large bone reference modelFDMLarge volume, detail tolerance is forgiving, PLA is fine and cheap at this scale Dental arch / crown prep study modelResin (MSLA)Sub-millimeter occlusal detail and undercuts need resin-level resolution Vascular or airway models (hollow, thin-walled)Resin, flexible or clearThin-wall accuracy and the option of a see-through material for teaching Pre-surgical "rehearsal" model with internal detailResin or multi-material FDMOften needs distinct colors/densities for different tissue typesFor dental work specifically, dimensional accuracy is the whole game — a crown prep model that's off by 0.2 mm can make a lab-fabricated crown not seat correctly. Run your printer's standard calibration (XY compensation, exposure test, whichever applies) before trusting any dental output, and validate a new workflow against a known physical model before relying on it for anything that affects an actual patient's care.
Safety, Privacy, and Liability
- Strip PHI (protected health information) from every DICOM file before it leaves a secure environment — name, DOB, MRN, and institution are all embedded in the header by default.
- This is not medical advice and these are not medical devices. A 3D print made on a hobby printer from a segmented scan is a visualization and education aid, not a validated surgical instrument, unless produced through a cleared regulatory pathway with biocompatible, traceable materials.
- Resin handling applies all the usual rules — gloves, IPA wash, full UV cure, and uncured resin treated as the irritant it is, same as any other SLA work.
- If a clinician is involved in the request, loop them in on material and accuracy limitations explicitly — don't let a hobby print get treated as more authoritative than it is.
Once the pipeline is set up, turning a scan into a model is mostly patience in Segment Editor rather than any exotic hardware requirement — the same resin or FDM printer you already use for everything else handles the output just fine. The place to spend your care is upstream, in getting the segmentation clean and the privacy handling right, not in the printing itself.