A peer-reviewed multi-institution feasibility study published on 8 October 2026 examined how patient-specific 3D-printed models can fit into embolisation planning for pulmonary arteriovenous malformations, or PAVMs. It covered eight lesions in six patients treated at two hereditary haemorrhagic telangiectasia centres. Every model was produced and used by the interventional radiologists, and angiographically defined technical success was achieved for all eight lesions.

What was modelled, and why is the procedure difficult?

A PAVM creates a direct connection between a pulmonary artery and vein. During embolisation, the operator must understand the number, size and spatial orientation of feeding arteries, then stop flow with plugs or coils. Four of the eight lesions were simple and four complex; six were de novo and two were recurrent after earlier treatment. Mean patient age was 50.3 years, with a range from 16 to 87. All five patients who underwent genetic testing were positive for an ENG mutation.

The centres did not apply the same threshold for ordering a model. One reserved printing for anatomy expected to be unusually difficult, while the other also printed typical cases because the technology was more accessible and less expensive locally. This matters methodologically: the series does not represent a random cross-section of PAVMs but two distinct clinical workflows.

From CT data to a physical model

The teams segmented thin-slice contrast-enhanced CT data in Materialise Mimics Medical. Geometry was prepared in 3-matic or Geomagic Freeform, and a radiologist reviewed segmentation accuracy. One-to-one scale models were built on a Stratasys J750 material-jetting system or an HP MJF 580. One centre used both rigid and flexible materials to represent different tissue behaviour; the other produced rigid models.

A strength of the workflow is clinical review of the digital anatomy before manufacture. That does not make the object an independent diagnostic device: segmentation, mesh processing, printing and post-processing can each introduce error. The study reported a numerical physician score only for three models at one centre; the operator rated all three 5/5 for agreement with anatomy encountered during the procedure.

How the models affected planning

All six models were used before the procedure and as an intraprocedural reference. The authors report that operators gained a better understanding of vascular anatomy and the preferred approach. In one complex case, an earlier embolisation attempt without a model had failed. The printed anatomy was later used to heat-shape a catheter, enabling access to and occlusion of the feeding vessel. Two other patients required staged treatment because of lesion size and complexity.

The outcome definition was narrow: technical success meant angiographic cessation of flow after embolisation. It was achieved for all eight PAVMs, but a small uncontrolled series cannot establish that printing caused the success. Procedure time, fluoroscopy time, radiation dose, cost and long-term clinical outcomes were not measured. Models were also used to educate five patients, but comprehension and satisfaction were not quantified.

Verified fact, interpretation and limitation

Verified facts are that six models were made, all eight treated lesions met the predefined technical endpoint, and a catheter shaped with model assistance succeeded in one previously unsuccessful case. The claim that models improved anatomical understanding and strategy is the clinicians' interpretation and is based mainly on operator reporting.

The retrospective study includes only six patients, has no no-model comparison group and obtained only qualitative feedback at one centre. It is therefore feasibility evidence, not a clinical efficacy trial. A useful next step would be a larger prospective comparison with predefined endpoints: procedure and fluoroscopy time, radiation exposure, device use, cost, retreatment and patient understanding.

Editorial conclusion

PAVM is a plausible target for patient-specific physical models because clinicians must navigate a small, spatially complex vascular network. The most useful result is not the 100% figure in isolation, but that two institutions integrated the workflow using different printing platforms. Wider adoption still needs reproducible segmentation, dimensional checks, documented production time and cost, and comparative clinical evidence. The print is an adjunct; it does not replace imaging, interventional judgement or a controlled medical-device workflow.