AI in medicine and 3D

From image to object

Turning CT and MRI into printable anatomical models with as few manual decisions as possible.

The step that dominates both time and quality is segmentation. Today it is done by hand, slice by slice, by someone who knows where density misleads: the diploë that vanishes as the threshold rises, the haematoma touching the cortex, the foramen that morphology silently closes. That is exactly where a trained model has work to do.

The lab already runs the full pipeline for skull, spine, brain (cortical reconstruction with FreeSurfer) and CT-angiography vasculature, and applies it beyond the clinic too: reproducing hominin skulls to teach human evolution, and digital morphometry on models from open academic repositories.

• In development: an auto-segmenter for 3D printing, with mandatory human review before export.

From image to decision

Scoring instruments that turn an imaging finding into a justifiable — and auditable — course of action.

Many of the scales neurosurgery relies on were built decades ago, on imaging technology that is no longer ours. The lab develops its own instruments through a systematic workflow: domain definition, weighting and cohort validation before any clinical use.

None replaces the team's judgement; all exist to make that judgement explicit.

From model to theatre

Bringing planning into the operation itself without depending on a proprietary platform.

Craniotomy planning, spinal instrumentation planning and haematoma volumetry already run as lab tools in the browser.

The next front is navigation: augmented-reality neuronavigation with off-the-shelf devices for trauma and emergency haematoma, a simple low-cost neuronavigator, and a 3D-printed robotic arm for spine and frameless stereotaxy. This front is in development and has no clinical use yet.