Deep Learning Based Surface Classification of Functionalized Polymer Coatings

Low-technology characterization of material surfaces poses a challenge of significant importance for many scientific fields such as medical implants, biosensors, and regenerative medicine. Simple, fast, and scalable surface analysis methods that can be applied to a wide range of functionalized polymer coatings would thus constitute a major scientific and technological advance. In this work, we studied stain patterns formed by depositing a defined protein solution onto various polymer surfaces. The images of the resulting drying droplet patterns were captured by polarized light microscopy and analyzed by a deep-learning neural network. In this proof-of-concept study, we used chemical vapor deposition polymerization to deposit ten structurally distinct polymer coatings that share an identical polymer backbone, but differ in their functional groups. Despite the relatively minute differences in their chemical structure, the CNN classification of the stain patterns was highly reproducible. Across all different polymers, the overall classification accuracy of the CNN was 96%. When challenging the CNN with images from an unknown polymer coating, i.e., poly[(4-bromo-p-xylylene)-co-(p-xylylene)], these surfaces were classified as halogenated or pseudohalogenated coatings with 95% accuracy. These findings confirm that the scope of surfaces that can be analyzed with this approach goes beyond polymer coatings already known to the CNN through the training procedure and validates the method as a simple, yet versatile surface analysis tool.