Integrating AI with liquid-phase transmission microscopy for dynamic assembly of nanoparticles

Abstract

Understanding solution-state particle dynamics has driven significant importance in comprehending their mutual interactions, despite their rapid Brownian motion and unpredictable interaction with solvent. While liquid-phase transmission electron microscopy (LP-TEM) enables direct visualization, quantitative analysis is limited by low contrast and background effects. We present a framework integrating deep learning with in-situ TEM for enhanced visualization. Our approach combines U-Net architecture for LP-TEM video binarization with ImageJ trajectory analysis. This enables precise particle tracking with four-dimensional spatiotemporal information and enhanced contrast. The framework shows efficacy in analyzing molecular interactions in liquid environments, improving conventional methods. This integration provides a platform for investigating nanoscale solution-state dynamics, advancing understanding of self-assembly mechanisms and particle interactions.