Artificial neural network-guided optimization of Kagome bone regeneration scaffolds based on asymptotic expansion homogenization results and its validation via additive manufacturing

Developing polymer-based bone regeneration scaffolds remains a challenge due to the difficulty of replicating the mechanical behavior of natural bone. Among various structural designs, the three-dimensional Kagome structure stands out for its excellent stiffness to weight ratio and efficient load distribution. However, the optimal design parameters that balance mechanical stiffness with pore size and porosity have not been fully explored. This study presents an artificial intelligence (AI) approach for bone regeneration scaffold design optimization. An artificial neural network (ANN) model was developed and trained using a precomputed dataset of Kagome scaffold geometries and their mechanical properties. The ANN model enables rapid prediction of normalized effective elastic stiffness ( ) without requiring iterative numerical simulations, thereby significantly reducing computational cost and accelerating the design optimization process. The ANN model was used to explore and optimize key design parameters of Kagome scaffolds, including pore size, porosity, column length ( ), strand diameter (D), frame column angle (θ), and frame length ( ). Their effects on were analyzed through the ANN model and validated using the Asymptotic Expansion Homogenization (AEH) method. The result of this approach is the Highly Optimized Effective Stiffness (HOES) Kagome bone regeneration scaffold. This scaffold was fabricated using polycaprolactone (PCL) through precision extrusion-based 3D printing. Mechanical testing revealed a 43.2% increase in compared to a conventional Kagome bone regeneration scaffold. The close agreement between ANN predictions, numerical simulations, and experimental results confirms the effectiveness of the proposed optimization strategy. In addition, in-vitro cell proliferation assays confirmed the biocompatibility of scaffolds.