Machine learning-driven multi-objective optimization of Gyroid bioglass scaffolds for site-specific biomechanical adaptation.

The clinical translation of scaffold-driven bone regeneration is hindered by an inherent topological paradox between mechanical integrity and mass transport, leading to biomechanical mismatches at heterogeneous anatomical sites. This study proposes an integrated "design-to-fabrication" protocol to address this bottleneck. A high-fidelity database of specific surface area, compressive strength and permeability was generated for 56 Gyroid scaffolds via geometric calculation, finite element analysis (FEA), and computational fluid dynamics (CFD). Support vector regression (SVR) models accurately mapped geometric parameters to these performance metrics (R2 > 0.98), driving NSGA-II multi-objective optimization to explore design trade-offs. Optimized scaffolds were fabricated via digital light processing (DLP) using an inverse compensation strategy. From the resulting Pareto frontier, two scaffolds meeting two distinct clinical requirements were screened out: a mechanics-prioritized scaffold (Point 1) achieving 25.57 MPa compressive strength suitable for load-bearing regions, and a transport-prioritized scaffold (Point 2) with enhanced permeability and uniform flow favorable for vascularization. The compensation strategy constrained fabrication errors to below 3%. This protocol integrates machine learning, multi-objective optimization, and high-precision additive manufacturing, offering a scalable pathway to develop bone scaffolds that rapidly meet specific clinical needs.