Inverse design of anisotropic bone scaffold based on machine learning and regenerative genetic algorithm

Area of Science:

• Biomaterials Engineering
• Computational Mechanics
• Artificial Intelligence in Medicine

Background:

• Triply periodic minimal surfaces (TPMS) offer structural advantages for bone scaffold design.
• Current TPMS scaffold design is limited to a forward process, predicting properties from structure.
• An inverse design approach, tailoring structure to mechanical properties, is needed.

Purpose of the Study:

• To develop an inverse design model for anisotropic bone scaffolds using TPMS structures.
• To establish a method for designing bone scaffolds based on target mechanical properties.
• To improve the efficiency and control of bone scaffold design.

Main Methods:

• A novel inverse design model combining machine learning (backpropagation neural network - BPNN) and a regenerative genetic algorithm (RGA) was developed.
• Finite element (FE) analysis was used to generate a BPNN dataset, correlating microstructural parameters with bone's elastic matrix.
• Bone's anisotropic mechanical properties were mimicked by adjusting cell density in different directions.

Main Results:

• The BPNN-RGA model accurately predicted the elasticity matrix of inverse-designed TPMS bone scaffolds.
• Average errors were below 3.00% for three mechanical performance targets and approximately 5.00% for six targets.
• The model demonstrated high design efficiency compared to traditional optimization methods.

Conclusions:

• Combining machine learning and optimization techniques enables the inverse design of anisotropic TPMS bone scaffolds with specific mechanical properties.
• The BPNN-RGA model offers a more efficient and controllable design process for bone scaffolds.
• This approach holds significant potential for advancing bone tissue engineering and regenerative medicine.