Data-driven inverse design of novel spinodoid bone scaffolds with highly matched mechanical properties in three orthogonal directions

• Data-driven inverse design of novel spinodoid bone scaffolds was performed. • A new approach combining genetic algorithm and deep learning was adopted. • Mechanical properties of the scaffolds were highly matched to those of the target bone tissues. • Anisotropies of the scaffolds were also close to those of the target bone tissues. Bone scaffolds are widely used in orthopedics for repairing bone defects and promoting bone regeneration. However, the issue of stress shielding caused by an excessive elastic modulus and mismatched anisotropy in bone scaffolds remains unresolved. Therefore, it is essential to design novel bone scaffolds with mechanical properties that closely match those of human bone. In this study, a novel data-driven inverse design framework was proposed to design spinodoid bone scaffolds by combining a back propagation neural network with a genetic algorithm. For spinodoid bone scaffold type Ⅰ, compared to the target human bone, the relative errors on the nine independent constants of elasticity matrix ranged from 0.090% to 6.444%. Similarly, for spinodoid bone scaffold type Ⅱ, the relative errors ranged from 0.000% to 7.084%. Both the elastic constants and the anisotropies of the novel bone scaffolds were highly matched to those of the target bone tissues in all the three orthogonal directions. Moreover, the results from data-driven inverse design were compared with those obtained from finite element analyses and validated by experimental tests. The proposed data-driven inverse design of spinodoid structures holds promise for further exploration in tissue engineering and other scientific fields.