Abstract

Scanning electron microscopy (SEM) is widely used for characterizing the surface morphology of nanostructured biomaterials, drug delivery systems, and biomedical devices at micro- and nanoscales. Despite its ability to provide high-resolution structural information, the interpretation of SEM images remains largely dependent on manual analysis, which is time-consuming and susceptible to operator-dependent variability. Recent advances in artificial intelligence, particularly deep learning approaches based on convolutional neural networks (CNNs), have created new opportunities for automated and reproducible analysis of SEM data. This review critically examines recent studies that apply deep learning techniques to SEM image analysis in nanomedicine and biomedical materials research. Core principles of CNN-based image analysis are briefly introduced, followed by an overview of commonly investigated morphological features and classification tasks. The review discusses reported strategies for dataset construction, image preprocessing, model training, and performance evaluation, highlighting both methodological trends and recurring limitations in the literature. Key challenges, including limited dataset sizes, non-independent data sampling, variability arising from imaging conditions, and issues related to model interpretability and generalizability, are also addressed. Finally, the review outlines future directions for improving the robustness and translational relevance of AI-assisted SEM analysis, with particular emphasis on reproducibility, validation across instruments, and potential applications in preclinical research and quality control.

Keywords

scanning electron microscopy, deep learning, convolutional neural networks, nanomedicine, biomedical materials, image analysis