Developing the Computational Image Processing Method for Quantitative Analysis of Nanopore Structure Obtained from HS-AFM (AFMnanoQ)

High-Speed Atomic Force Microscopy (HS-AFM) enables nanoscale imaging of biological structures with exceptional spatial (1 nm in the x-y plane; ~0.1 nm in the z-direction) and temporal resolution (~20 ms per frame). HS-AFM encodes three-dimensional (3D) information within a two-dimensional (2D) format, where the lateral dimensions (x, y) of structures correspond to spatial positioning in the image, while height (z) information is embedded in pixel intensity. This unique data structure presents significant challenges in segmentation and morphological analysis, requiring specialized computational approaches. To overcome these limitations, we develop “AFMnanoQ”, a feature-driven computational framework for segmentation and morphological measurement of HS-AFM data. Our method operates independently of labeled training data, making it robust to data scarcity while simultaneously serving as a powerful tool for generating high-quality labeled datasets for future deep-learning applications. We validate AFMnanoQ using both synthetic and experimental AFM/HS-AFM datasets, including semi-automatic analyses of the conformations and dynamics of alpha-hemolysin (αHL), a β-barrel pore-forming toxin (PFT) secreted by Staphylococcus aureus . Our method achieves competitive performance with deep-learning models while maintaining superior adaptability across diverse HS-AFM datasets. As the future perspective, we plan to further develop or integrate it with deep-learning models to enhance segmentation performance and to reconstruct 3D structures from experimental AFM images. This will leverage conformational libraries generated in this study, enabling cross-validation between the two methods and ultimately bridging the gap between feature-driven and data-driven approaches in AFM image analysis.