Enhancing image resolution of confocal fluorescence microscopy with deep learning

Abstract Super-resolution optical imaging is crucial to the study of cellular processes. Current super-resolution fluorescence microscopy restricted by need special fluorophores or sophisticated systems, long acquisition and computational times. In this work, we present a deep-learning-based technique confocal microscopy. We devise two-channel attention network (TCAN), which takes advantage both spatial representations frequency contents learn more precise mapping from low-resolution images high-resolution ones. This scheme robust against changes in pixel size setup, enabling optimal model generalize different modalities unseen training set. Our algorithm validated on diverse biological structures dual-color actin-microtubules, improving resolution ~ 230 nm 110 nm. Last but not least, demonstrate live-cell revealing detailed dynamic instability microtubules.