Not AI-assisted.
Not AI-enabled.
Implant, instruments, and intelligence, co-designed from the ground up — an AI-native approach to total knee arthroplasty, beginning with the bicruciate-retaining knee.
The knee that patients prefer most
is the knee that surgeons use least.
A million Americans receive a new knee each year. Nearly all of them lose their anterior cruciate ligament — the structure that makes a knee feel like a knee. The bicruciate-retaining (BCR) procedure preserves it, along with the posterior cruciate. It is what patients want, and what surgeons can almost never deliver.
Three generations of bicruciate-retaining knees have tried, over fifty years. None have scaled. The failure mode is the same in every generation: BCR is so mechanically sensitive that the margin for surgical error is smaller than what conventional instrumentation can guarantee.
Multiple attempts across four decades. Ligament sparing was correct in principle; surgical precision was too manual to scale.
Modern materials, conventional instrumentation. Withdrawn after three years — component placement tolerances could not be met consistently.
The last available BCR in the U.S. Withdrawn December 2025 in portfolio rationalization. Zero BCR implants currently on market.
The knee is not a hinge. Its cruciate ligaments form a crossed four-bar linkage whose instantaneous center of rotation migrates through flexion — a polycentric motion with no single pivot to align to. Below: the same knee with the ACL preserved, and with it resected. The mathematics of alignment are different in the two cases.
BCR is so mechanically complex
it fundamentally requires AI to be performed reliably.
Every patient is different.
Every plan starts from the difference.
The implant is standardized. The surgeon is constant. The anatomy is not. Cartan begins each case from the patient's own scan — their own bone, their own ligament geometry, their own kinematics — and plans around them, not around an average.
Six cadaveric knees from the University of Denver Natural Knee Dataset. Each one is its own patient; each one would need its own plan.
- Bone and cartilage surfaces, from CT and MR, within the patient's coordinate frame.
- Cruciate and collateral footprints — where the ligaments attach, not where they usually do.
- A patient-specific kinematic model that drives the 50-year simulation, the surgical plan, and the intra-operative guidance.
The six specimens shown are a sample of the University of Denver Natural Knee Dataset (Rullkoetter, Laz, Shelburne et al.). On Cartan's production planner, the same pipeline runs from an incoming patient's MRI and CT — the digital twin is built from their scan, not selected from a library.
Fifty years of wear,
simulated before the first incision.
Cartan's digital twin exposes the long-horizon mechanics of a procedure before it happens. Contact pressure evolves as a scalar field on the tibial cartilage; wear, fracture, and loosening hazards accrue on parametric curves. The same patient, scrubbable across half a century.
Curves are illustrative, tuned to published arthroplasty trends for readability — not clinical predictions. The same shape-space and simulation pipeline, driven by real patient-specific input, will feed Cartan's pre-operative planning interface.
Align the implant
to the patient's moving frame.
Chasles' theorem says every motion is a screw about an instantaneous axis. Sweep flexion and that axis traces a ruled surface — the axode — a compact signature of native knee motion. A faithful bicruciate-retaining procedure has to restore it. Not by matching surfaces: implant geometry and native anatomy obey different kinematic constraints, and the difficulty of mapping between them is exactly why BCR has historically failed to achieve mainstream utilization.
At each instant during flexion, the tibiofemoral joint screws about a single line in space — its instantaneous helical axis. The one-parameter family of those lines sweeps a ruled surface: the axode.
Restoring the native axode with a bicruciate-retaining implant is not a resurfacing problem. The surfaces of the implant do not behave kinematically as the surfaces of the native joint; reproducing the shape of the knee does not reproduce its motion. The surgical plan has to solve the harder version — which is why BCR has resisted three generations of attempts, and why Cartan's planner is the first thing built around it.
Every procedure
makes every future procedure better.
Each case Cartan runs leaves a trace. Pre-op imaging, the surgical plan, the intra-op execution, and the post-op outcome together place the patient in a learned latent space — a map of everyone Cartan has ever treated. As cases accumulate, the model's prediction for the next patient sharpens.
- Pre-operative outcome prediction for the next patient.
- Alignment targets specific to each anatomical cluster.
- Early detection of rare failure modes the first procedures never saw.
Cartan's platform is architected against the FDA's Predetermined Change Control Plan (PCCP) framework — the flywheel can keep turning without a new 510(k) each revolution. The axes shown here are synthetic; the structure is not.