Machine Learning Optimized 3D Scaffold Pores For Bone Regeneration

To identify the optimal pore size (50–400 μm) and porosity (30–90%) of 3D biomaterial scaffolds for osteogenic differentiation using integrated experimental and machine learning approaches. Nine scaffold groups with varying pore sizes/porosities were tested. Osteoblast proliferation (CCK-8/EdU), viability (Live/Dead staining), differentiation (ALP/RUNX2/BMP2), and mineralization (Von Kossa) were evaluated over 21 days. Machine learning models—including multiple linear regression (quantitative relationships), random forest (feature importance), SVM/KNN classifiers (high-performance group prediction), and PCA (data pattern visualization)—were applied to analyze scaffold parameters and osteogenic outcomes. Scaffolds with medium pore sizes (150–200 μm) and porosity (50–70%) exhibited 40–60% higher cell proliferation and peak ALP activity (12.8±1.3 U/mg). Machine learning identified porosity as the primary determinant (feature weight: 0.62), where medium porosity optimized nutrient transport, while medium pore size enhanced mineralization (35% increase in calcium nodule density). SVM/KNN classifiers achieved >85% accuracy in predicting optimal parameters (150–200 μm/50–70%). This machine learning-guided framework efficiently identifies scaffold parameters that maximize bone regeneration, with 150–200 μm pore size and 50–70% porosity demonstrating superior performance. The approach provides a data-driven strategy for developing personalized bone repair materials.