AI Prototype Designs Custom Foot Orthoses with Semantic-Physics Alignment

Translating high-level clinical prescriptions into patient-specific foot orthoses (FOs) is challenging due to a "semantic-physical misalignment," where clinical intent doesn't directly map to 3D geometric parameters. Current design workflows are manual and lack instantaneous biomechanical validation. Researchers present TANS-FO, a modular research prototype for the closed-loop computational design automation of customized FOs. It features a Text-Aligned Neural Surrogate (TANS) that projects clinical-text embeddings onto a continuous lattice-density field. A Graph Neural Network (GNN) surrogate is integrated to predict plantar stress in real-time, effectively replacing time-consuming Finite Element Analysis (FEA). Anchored on the PicoFoot-5K anthropometric database, the framework synthesizes manufacturing-ready lattice insoles within minutes. The GNN surrogate shows high agreement with reference solvers (R^2 = 0.94). The system achieved a 34.7% peak-pressure reduction over parametric CAD and a fit error of 0.42 mm. While preliminary observational data suggests short-term comfort improvement, the authors explicitly state this is not clinical efficacy evidence. This prototype demonstrates a significant step towards automated, patient-specific medical device design.