ML-Enabled Optimisation Framework for High-Performance Tissue Scaffold Design
Please login to view abstract download link
Machine learning (ML) enabled modelling and design of tissue scaffolds is emerging as a powerful approach in regenerative medicine [1,2], where long-term bone ingrowth is a key performance objective. This study presents a novel computational optimisation framework for patient-specific scaffold design that maximises time-dependent tissue ingrowth. The proposed method integrates ML within a dynamic optimisation framework: two neural networks are embedded in a bone growth model, and a third network is coupled with a genetic algorithm to accelerate the optimisation process [3]. The approach is demonstrated using a sheep mandible reconstruction case involving a critical-size bone defect. The finite element model is experimentally validated through mechanical testing of a tailored 3D-printed PEK scaffold. Three optimisation strategies — uniform, lateral-gradient, and vertical-gradient designs — are evaluated against an empirical design under identical biomechanical conditions. The optimised scaffold achieves an 18.5% improvement in predicted long-term bone ingrowth, driven by fine-tuned lattice strut distributions in gradient regions. The proposed ML-assisted, time-dependent optimisation framework establishes a new pathway for designing high-performance tissue scaffolds with improved regenerative outcomes. REFERENCES [1] Wu et al (2024) Dynamic optimisation for graded tissue scaffolds using machine learning techniques. Computer Methods in Applied Mechanics and Engineering 425, 116911. [2] Wu et al (2021) A machine learning-based multiscale model to predict bone formation in scaffolds. Nature Computational Science 1 (8), 532-541. [3] Wu et al (2021) Machine learning-based design for additive manufacturing in biomedical engineering. International Journal of Mechanical Sciences 266, 108828.
