Machine Learning-Assisted Evaluation of Topography-Dependent Cellular Responses in Meniscus-Mimetic Piezoelectric...

Electrospun nanofibrous scaffolds provide topographical cues that regulate cellular behavior; however, relationships between scaffold design and multidimensional biological responses remain difficult to define because structural and material parameters act simultaneously. Here, we fabricated meniscus-mimetic piezoelectric nanofibrous scaffolds with different fiber architectures and compositions and systematically evaluated their cellular responses using machine learning (ML). Poly(vinylidene fluoride) (PVDF)-based scaffolds with random, aligned, and radial fiber architectures were prepared, including composite scaffolds containing barium titanate (BaTiO₃) and multi-walled carbon nanotubes (MWCNTs). Scaffold morphology and fiber organization were quantitatively characterized, and mesenchymal stem cells (MSCs) cultured on the scaffolds were analyzed by DAPI, F-actin, and SRY-box transcription factor 9 (SOX9) fluorescence imaging. Quantitative image analysis revealed scaffold-dependent differences in cellularity, nuclear morphology and orientation, cytoskeletal organization, and SOX9-associated phenotype, demonstrating distinct cellular responses to fiber topography and material composition. Scaffold and cellular parameters were subsequently integrated into a quantitative dataset to develop and compare multiple ML models for modeling scaffold-dependent cellular responses. The ML framework captured relationships between scaffold characteristics and multidimensional biological outcomes. These findings establish an experimental-computational approach for linking electrospun scaffold design with cellular behavior and provide a data-driven framework for the rational evaluation and optimization of meniscus-mimetic piezoelectric scaffolds.