Predicting environmental pollutant concentrations via cell image-derived damage features using a hybrid model

Introduction Effective detection of environmental pollution relies on reliable methods for determining pollutant concentrations, with cellular damage reflecting pollutant toxicity as a vital detection tool. This study presents a novel quantitative method for predicting environmental pollutant concentrations using cell images and a hybrid model. Methods The approach processes conventional optical microscope images by extracting grayscale statistical features and constructing a hybrid predictive framework that integrates stepwise regression for feature selection and multilayer perceptron for nonlinear modeling, enabling accurate mapping from image-based damage features to pollutant concentrations. Results Experiments show that the model performs consistently well across five cell types: HeLa, A549, HUVEC, PC12, and HaCaT. For example, it achieves an R 2 of 0.9911 on the HeLa test set, demonstrating strong generalization ability and robustness. Discussion The method does not require expensive equipment or complex sample preparation, offering an innovative, rapid, and low‐cost solution for monitoring environmental pollutant concentrations.