Generative AI for real-time super-resolution images of cellular microtubules

Abstract The study of living cellular structures, long constrained by the optical diffraction limit, has been revolutionized by Super-Resolution Microscopy, which enables visualization with unprecedented resolution. This gain in spatial resolution, however, is achieved at the expense of protracted acquisition times, making these techniques ill-suited for imaging highly dynamic biological processes. Here, we propose the adoption of supervised Deep Learning models, such as the Enhanced Super-Resolution Generative Adversarial Network and the Dense Residual Connected Transformer, to mitigate the inherent trade-off between spatial and temporal resolution that limits live-cell imaging. Specifically, this study focuses on generating super-resolved images of microtubules, simulating acquisitions from Stochastic Optical Reconstruction Microscopy. Once trained, the models can generate a super-resolution image in a matter of seconds, starting solely from a diffraction-limited image obtained via conventional fluorescence microscopy.