A patient-specific anatomical model does not become clinically suitable merely because it resembles a CT rendering. Imaging, segmentation, mesh repair, scaling, printing and post-processing can all alter geometry. Validation therefore is not the production of one accuracy number; it is evidence that the whole workflow is reliable enough for a predefined clinical purpose.

This guide focuses on anatomical models and imaging or radiotherapy phantoms. It does not replace local quality-management, medical-device, radiation-safety or clinical rules. FDA guidance, RSNA consensus work and the 2026 AAPM TG-336 report share a central principle: design requirements, the software chain, material, process and finished object must be controlled as one system.

1. Intended use defines tolerance

First state what the model is for. Patient communication may require recognisable large structures; surgical planning may depend on the distance between a tumour and vessel; device rehearsal may require lumen diameter, wall compliance and deformation under pressure. The same print can be acceptable for one purpose and unsuitable for another.

Requirements should list critical anatomy, permitted geometric error, mechanical or optical properties, sterilisation where relevant, identification, usable life and the decision the model supports. Tolerance should follow clinical risk and source-image resolution, not the printer brochure.

2. Imaging is part of manufacturing

Slice thickness, reconstruction kernel, contrast, motion and partial-volume effects constrain visible boundaries. A printer cannot faithfully recover anatomy absent from the source data. The imaging protocol and DICOM series must therefore be tied to the manufacturing record.

Segmentation records should capture software and version, thresholds, manual edits, operator and approval. Smoothing, hole closure and wall thickening may make a model printable while changing anatomy. Modified regions and their rationale need traceability.

3. Digital review before printing

After technical mesh repair, critical cross-sections should be checked against the original images. Overlaying the segmented surface on CT or MR slices is more informative than viewing an attractive 3D rendering. A second qualified reviewer can reduce operator dependence in high-risk cases.

File units, coordinate system, scale and orientation must be fixed before slicing. STL does not reliably carry units or clinical metadata, so a filename is weak identification. Use a version identifier, a privacy-preserving case number and a controlled export.

4. Qualifying the print process

Installation, operational and performance qualification are distinct levels. A regular test artefact should reveal scale error, flatness, hole and wall reproducibility and position effects across the build volume. Material lot, profile, orientation, supports, build location, post-cure or heat treatment and cleaning are controlled variables.

FDA's process description distinguishes characteristics that can be inspected individually from those requiring prior process validation because testing would be destructive or impractical. External dimensions can be measured, but the strength of every internal wall may not be. A stable process window and change control are therefore as important as final inspection.

5. Geometric verification of the finished model

Calibrated callipers or gauges may suit simple external dimensions. Optical scanning and comparison with the approved digital model provide a dense deviation map for complex surfaces. Surface scanners cannot see internal channels; CT or other volumetric imaging may be required.

AAPM TG-336 stresses that every measurement method has limitations. A flexible model can deform under its own weight, and a pressurised vessel model may expand significantly depending on material, temperature and diameter. Measurement setup and mechanical state must reflect intended use.

Do not report only mean surface error. Measure maximum deviation, diameters, distances and wall thickness at critical locations. Measurement uncertainty belongs in the decision; a result close to a tolerance limit cannot automatically pass.

6. Material, colour and function

Geometric accuracy does not prove functional suitability. A vascular rehearsal model may require compliance, friction, puncture resistance and pressure behaviour. Colour coding can aid surgical planning but must not obscure anatomy. A radiotherapy phantom needs separate validation of attenuation and homogeneity.

Evaluate material in the finished, post-processed product. A generic resin or powder data sheet does not prove biological, mechanical or imaging performance for a specific model. Cleaning, washing, curing and ageing can change properties.

7. Acceptance and release

A practical release record includes case ID, source-image series, segmentation and mesh versions, printer and material IDs, profile revision, post-processing, visual inspection, critical measurements, result and approver. The label should state purpose, orientation, manufacture date and important limitations.

A nonconformance should trigger more than a reprint. Determine whether the error arose in imaging, segmentation, export, slicing, machine, material or post-processing. This turns QA into a learning system and prevents recurrence.

8. Periodic checks and change control

Use a recurring reference phantom to monitor machine and workflow stability. A new software version, material, printer, profile, curing unit, cleaning method or major segmentation algorithm may require risk-based revalidation. Frequency should follow use, failures, maintenance and clinical risk.

Verified fact, professional guidance and conclusion

FDA describes design, software workflow, material controls, printing, post-processing, validation and testing as separate elements of the additive medical-device process. RSNA and AAPM professional guidance extends QA from imaging to verification of the finished model. Those documents do not prove that any particular model improves clinical outcomes.

The editorial conclusion is straightforward: an anatomical model is trustworthy when purpose-derived requirements, complete digital and physical provenance, appropriate measurement and an approval decision form one documented chain. A visually convincing print or generic machine-accuracy claim cannot replace that chain.