Abstract
AbstractRecent advances with deep neural networks have shown the feasibility of acquiring brightfield images with transmitted light and applying in-silico labeling to predict fluorescent images. We have developed a novel in-silico labeling method based on a generative adversarial network and outperforms the state-of-the-art Unet method in generating realistic fluorescent images and quantitatively recapitulating real staining signals, as demonstrated in a complex co-culture myelination assay. Furthermore, we have performed the assay in live mode with multiple kinetic points, applied in-silico labeling to predict fluorescent images from brightfield and quantified the kinetic phenotypic changes. Thus, the proposed approach provides a potential tool to study the kinetics of cellular phenotypic changes with brightfield imaging.
Publisher
Cold Spring Harbor Laboratory