Deep learning-based single-shot structured illumination microscopy

We report a deep learning-based structured illumination microscopy (SIM) method, which can reconstruct a super-resolution (SR) image using only one frame structured illumination image. Generative adversative networks (GANs) and deformation of U-Net (DU-Net) are employed to perform the task. GANs are trained to generate other structured illumination images by feeding a single structured illumination image, and DU-Net is trained to reconstruct the super-resolution image. The results of experiments and simulations demonstrate that the SR image could be reconstructed from one frame structured illumination image. Importantly, it can greatly reduce phototoxicity and photobleaching.

Automated summary

This research presents a deep learning-based method for structured illumination microscopy (SIM) that can reconstruct super-resolution (SR) images using only one frame of structured illumination image. Generative adversative networks (GANs) and deformation of U-Net (DU-Net) are utilized for the task. Experimental and simulation results demonstrate the ability to reconstruct SR images from a single frame of structured illumination image, significantly reducing phototoxicity and photobleaching.