Traditional custom orthotics manufacturing has long struggled with the "semantic-physical gap." This is a polite way of saying that a doctor’s handwritten medical prescription often bears little correlation to the specific geometric parameters of a 3D model in a CAD system. Researchers from Mokwon University and Shanghai AI Generative Design have proposed a solution with the TANS-FO prototype. The system effectively translates clinical text into a print-ready 3D structure, removing the engineer from the production chain.
The project’s technological core is a Text-Aligned Neural Surrogate (TANS) that maps clinical data embeddings directly onto a lattice density field. To avoid waiting hours for heavy finite element analysis (FEA) results, Rui Wang and his colleagues implemented a Graph Neural Network (GNN). According to the report, the GNN predictor achieves an R2 accuracy of 0.94 compared to the professional Abaqus solver, providing physically accurate foot pressure forecasts in real time.
Key highlights of TANS-FO technology
PicoFoot-5K Integration: The model was trained on a database of over five thousand patients to handle complex anatomical variations. Graph Neural Networks: GNNs replace time-consuming mathematical simulations with instant load distribution forecasts. Precision Fitting: The average model generation error does not exceed 0.42 mm.
Practical trials with men aged 18–40 showed that the system reduces peak pressure 34.7% more effectively than standard parametric CAD templates.
For the medical hardware industry, this case represents a shift from manual drafting to intent-based design. We are seeing a rare example of AI doing more than just generating images; it is bridging medical semantics with real-world physics and the manufacturing cycle. While the project awaits regulatory approval, it already functions as a robust design support tool capable of drastically reducing skilled labor costs and eliminating human error in prescription interpretation.
Topic: business_cases