A neural network–based particle automated topography classification (PATC) model

To rigorously assess whether the engineered particles reproduced the biophysical characteristics of natural counterparts (i.e., Smooth, Convex, or Concave topographies), a neural network–based particle automated topography classification (PATC) model was developed. A custom transformer-based classification framework was developed to evaluate image-derived representations based on Qwen2-VL-7B-Instruct vision backbone. This model can extract and learn the image characteristics of natural particles, and then classify engineered particles as the category of corresponding natural counterparts.