Statistical Characterization of the Morphologies of Nanoparticles through Machine Learning Based Electron Microscopy Image Analysis
Byoungsang Lee, Seokyoung Yoon, Jin Woong Lee, Jin Woong Lee, Yunchul Kim, Junhyuck Chang, Jaesub Yun, Jae Chul Ro, Jong-Seok Lee, Jong-Seok Lee, Jung Heon Lee, Jung Heon Lee
Abstract
Although transmission electron microscopy (TEM) may be one of the most efficient techniques available for studying the morphological characteristics of nanoparticles, analyzing them quantitatively in a statistical manner is exceedingly difficult. Herein, we report a method for mass-throughput analysis of the morphologies of nanoparticles by applying a genetic algorithm to an image analysis technique. The proposed method enables the analysis of over 150,000 nanoparticles with a high precision of 99.75% and a low false discovery rate of 0.25%. Furthermore, we clustered nanoparticles with similar morphological shapes into several groups for diverse statistical analyses. We determined that at least 1,500 nanoparticles are necessary to represent the total population of nanoparticles at a 95% credible interval. In addition, the number of TEM measurements and the average number of nanoparticles in each TEM image should be considered to ensure a satisfactory representation of nanoparticles using TEM images. Moreover, the statistical distribution of polydisperse nanoparticles plays a key role in accurately estimating their optical properties. We expect this method to become a powerful tool and aid in expanding nanoparticle-related research into the statistical domain for use in big data analysis.