OsteoFlow: Lyapunov-Guided Flow Distillation for Predicting Bone Remodeling after Mandibular Reconstruction

A research team from the University of British Columbia and University of Toronto has developed OsteoFlow, a novel AI framework that predicts how bone will remodel and heal over a full year following complex mandibular reconstruction surgery. The system addresses a critical clinical challenge: standard generative AI models struggle to maintain anatomical consistency when forecasting biological processes over long time horizons. OsteoFlow's breakthrough comes from its Lyapunov-guided trajectory distillation approach, which distills continuous healing trajectories from registration-derived velocity fields rather than using conventional one-step prediction methods.

Unlike traditional models that might predict disjointed future states, OsteoFlow maintains geometric correspondence throughout the entire healing process by combining trajectory distillation with a resection-aware image loss function. This ensures the predicted bone growth follows physically plausible pathways while preserving the generative capacity needed for accurate CT scan synthesis. The model was rigorously evaluated on 344 paired regions of interest from clinical data, demonstrating a significant ~20% reduction in mean absolute error within the critical surgical resection zone compared to current state-of-the-art baselines.

The technical innovation lies in treating bone remodeling as a continuous flow process rather than a discrete prediction task. By learning from stationary velocity fields derived from medical image registration, OsteoFlow captures the underlying dynamics of bone healing more effectively than previous approaches. The framework's code has been made publicly available on GitHub, enabling further research and clinical validation. This work represents a meaningful step toward personalized surgical planning, where surgeons could simulate and compare potential long-term outcomes before making critical reconstruction decisions.

The clinical implications are substantial for maxillofacial surgery and orthopedics. Accurate prediction of bone remodeling could help surgeons optimize graft placement, anticipate complications, and set realistic patient expectations. While currently focused on mandibular reconstruction, the trajectory distillation methodology could potentially be adapted to other tissue regeneration predictions, opening new avenues for AI-assisted surgical planning across multiple medical specialties.