CT Scan 3D Printing: 7 Costly Mistakes to Avoid in 2025

• 1
  Symptom, likely cause, and what to do
• 2
  Treating DICOM data as if it were CAD geometry
• 3
  Letting automatic segmentation run unsupervised
• 4
  Exporting STL files with sloppy mesh hygiene
• 5
  Ignoring print orientation, supports and powder traps
• 6
  Choosing the wrong process, then skipping validation
• 7
  Picking a supplier with good machines and no process discipline
• 8
  A CT-to-part workflow that avoids all seven

A CT scan is a stack of gray values, not a surface. Every voxel carries a density reading plus noise, and the boundary between bone and air is a gradient, not a line. Teams that import DICOM into a slicer and expect a watertight model lose days. The scanner is rarely the problem. The data type is.

Slice pitch sets the smallest feature you can still see. A 0.5 mm cortical wall does not exist in a 1.5 mm reconstruction, and no amount of mesh smoothing will bring it back. For orbital floors or thin sinus walls, rescan at 0.5–1.0 mm slice thickness or accept that the model is an approximation.

Metal artifacts are the second trap. Dental amalgam, spinal screws and hip stems create beam hardening streaks that run across the whole reconstruction. Automatic thresholding reads those streaks as dense tissue. We have seen a cranial implant with a ridge that lined up exactly with a row of fillings.

Hounsfield units give you a starting threshold, not a final answer. Bone sits roughly 300 to 2,000 HU, soft tissue below 100 HU, but the correct cut point depends on the scanner, the kVp and the reconstruction kernel. Log those three values with the dataset. They decide whether your segmentation is repeatable.

• 1
  Check firstSlice pitch, kVp, kernel and pixel spacing before any segmentation.
• 2
  Red flagA single threshold value shared across two different scanners.

Region-growing and threshold tools are fast and mostly right. Mostly is the problem. They leak into adjacent structures through any gap in the boundary, and they stop early where contrast drops. A mandible with a thin condyle neck is a classic leak point.

The fix is boring and it works: run the automatic pass, then step through the volume slice by slice at the boundary. Check the condyles, the orbital floors, the sinus walls and any place where two bones touch. That review takes 30 to 60 minutes on a head scan. It saves a reprint.

For anything implantable or load-bearing, segment in two passes with different thresholds and compare the two masks. Where they disagree is where you need to look manually. This is how you catch a 0.3 mm boundary error before it becomes a fit failure.

Keep the mask as a separate file from the mesh. When a surgeon asks why the implant edge sits 0.4 mm proud, you want to show the mask that produced it, not just the STL.

• 1
  Two-pass compareDifferent thresholds, then overlay. Disagreement marks the review zones.
• 2
  Version controlDICOM, mask and mesh all carry the same revision number.

An STL has no units, no history and no tolerance record. It is a triangle list. Export it at a chord height of 0.5 mm and a 40 mm cranial curve becomes a visible polygon; export at 0.02 mm and the file is accurate but heavy. Pick the chord height from the smallest radius you must hold, not from the file size you prefer.

Non-manifold edges, flipped normals and intersecting shells all survive the export and then break the slicer. Most slicers will repair them silently, which is worse, because the repair can close a hole you intended. Run a mesh check and look at the reported error count before you send anything.

For CNC follow-up work on a machined implant or a guide, the mesh is only a reference. What matters is the STEP model rebuilt from it, because CAM needs real surfaces. A 0.1 mm stair step from the mesh becomes a chatter mark on the finished part.

Name your files so the revision is obvious: patient or part ID, date, chord height and a revision letter. Shops that handle medical work will ask for exactly that, and vague names slow the job down.

• 1
  Chord height0.02 mm for fine anatomy, 0.05 mm for large industrial parts.
• 2
  Mesh checkZero non-manifold edges, normals consistent, no self-intersections.

Orientation decides more than surface finish. On a surgical guide, the layer direction runs across the fitting surface, and layer lines on a fit surface show up as a 0.05 to 0.1 mm gap at the tooth. Rotate the part so layers run parallel to the contact face, then accept the longer build.

Supports leave marks wherever they touch. Put them on non-functional surfaces, and if none exist, add a sacrificial tab. On metal parts, supports also conduct heat away and can pull the part out of flatness on the first layers. Preheat and a thicker base plate help.

Hollow parts with trapped powder are a scrap risk. Un-sintered powder inside a closed shell will not come out, and it will show up on a CT inspection. Add drain holes of 3 to 5 mm at the lowest points of each cavity and verify them in the slice preview before the build starts.

Distortion is predictable. Long thin sections warp toward the heat source, and the first 3 mm of a build carry the most residual stress. If a part has a thin flange, orient it so the flange sits high in the build, away from the plate.

• 1
  Orientation ruleLayers parallel to the critical fit surface, supports off functional faces.
• 2
  Powder ruleEvery internal cavity gets two drain holes, minimum 3 mm.

Not every CT-derived part should be printed. A patient-specific metal implant that must hold ±0.05 mm and carry load is often better machined from Ti-6Al-4V on a 5-axis center, then finished, because the machined surface is uniform and the material properties are known. Print the model for the fit check, machine the part that goes into the body.

Anatomical models for surgical planning are the opposite case. They are large, organically shaped, single-use, and surface finish barely matters. Print them. We run these on 3-axis and 5-axis machines when a rigid, dimensionally stable model is needed, or on polymer printers when speed and cost dominate.

Validation is where projects quietly fail. A printed guide that fits a model but not the patient was never checked against the original scan. Overlay the finished part scan on the source DICOM and measure the deviation at the fit surfaces. Anything above 0.2 mm on a guide needs a conversation before surgery.

Write the acceptance criteria before the build. Which surfaces are critical, what tolerance applies, and what measurement method proves it. If that list does not exist, the part cannot be accepted or rejected on evidence.

• 1
  PrintAnatomical models, planning tools, large organic shapes, low load.
• 2
  MachineLoad-bearing implants, tight fits, known material properties, repeat runs.

Machine lists are easy to compare. Process discipline is not. Ask how the supplier handles the DICOM, who reviews the segmentation, and what happens when the mesh check reports errors. A shop that answers with a printer model instead of a workflow is a risk.

For medical work, ask for the quality system: ISO 13485 for the device side, ISO 9001 for general quality, and a traceable link from the raw data to the shipped part. GreatLight holds ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022, and we keep the DICOM, mask, mesh and inspection record tied to one job number.

Ask how inspection is done. On CT-derived parts, dimensional inspection on a CMM covers the machined features, but the freeform surfaces need a scan overlay. We inspect 100% before shipment and provide reports on request, with raw material check, in-process monitoring and final inspection in the sequence.

Finally, ask about confidentiality. Patient data is regulated in most markets. Uploads are secure and confidential, and an NDA is available on request. If a supplier treats a CT dataset like a casual photo, walk away.

• 1
  Ask forWorkflow document, not a machine brochure.
• 2
  Ask forTraceability from DICOM revision to shipped part number.