How our AI pipeline automates segmentation, landmarking, and deformity analysis end-to-end.
The planning bottleneck
Manual pre-operative planning for a corrective osteotomy typically takes between one and three hours of a surgeon's or engineer's time. Slices are reviewed by hand, bones are segmented semi-automatically, landmarks are placed one by one, and angles are measured in 2D projections that hide out-of-plane deformity. The result is accurate in experienced hands, but slow, hard to standardize, and difficult to scale across a department.
An end-to-end automated pipeline
SkeletonPlanner replaces that sequence with a single automated pipeline that runs from the moment a CT series is received.
- Segmentation — a cascaded deep-learning network separates femur, tibia, fibula, and patella at sub-voxel boundaries
- Landmarking — anatomical landmarks are predicted directly on the 3D surface rather than on 2D slices
- Axis reconstruction — mechanical and anatomical axes are derived from the landmark set
- Deformity analysis — HKA, MAD, mLDFA, MPTA, torsion, and length differences are computed automatically
- Plan generation — osteotomy planes, wedge geometry, and correction targets are simulated and exported
Where the time goes
On a standard lower-limb CT, segmentation takes roughly three minutes, landmarking and axis reconstruction under one minute, and full deformity analysis about two minutes. The surgeon then reviews and adjusts the plan — the step that genuinely requires clinical judgement — instead of spending that time on measurement mechanics.
Human in the loop, always
Every automated step is reviewable and editable. Landmarks can be nudged, planes can be repositioned, and the measurement set recalculates instantly. The output is a plan the surgeon owns, produced in minutes rather than hours.