Optimizing nanofiber morphology is essential for promoting osteoblast elongation and supporting bone regeneration. This study aimed to develop a machine-learning framework capable of predicting optimal scaffold architectures directly from scanning electron microscopy (SEM) images and chemical composition. A four-module pipeline was implemented, combining tile-based SEM preprocessing, Cellpose-based cell morphology extraction with edge correction, ensemble machine-learning models, and an end-to-end convolutional neural network (CNN). Cellular quality was quantified using an elongation-weighted metric to emphasize morphological maturity over cell number. The analysis revealed consistent structure–function relationships across samples, with Sample₅ achieving the highest quality score at the 72 h time point. Ensemble models reached an R2 of 0. 400, while the end-to-end CNN achieved an R2 of 0. 750, indicating that raw SEM texture provides additional predictive information beyond handcrafted features. Feature-importance analysis identified nonlinear MgO effects and synergistic interactions between MgO and gold nanoparticles as key determinants of cell morphology. These findings demonstrate that the integrated workflow can reliably identify morphology–chemistry combinations favorable for osteoblast performance and provide a foundation for data-driven scaffold optimization. The approach supports rational design of nanofibrous biomaterials and may facilitate future development of intelligent scaffolds for bone regeneration applications.
Radwan-Pragłowska et al. (2026) studied this question.