Surpassing the resolution limitation of structured illumination microscopy by an untrained neural network

Abstract

Structured illumination microscopy (SIM), as a flexible tool, has been widely applied to observing subcellular dynamics in live cells. It is noted, however, that SIM still encounters a problem with theoretical resolution limitation being only twice over wide-field microscopy, where imaging of finer biological structures and dynamics are significantly constrained. To surpass the resolution limitation of SIM, we developed an image postprocessing method to further improve the lateral resolution of SIM by an untrained neural network, i.e., deep resolution-enhanced SIM (DRE-SIM). DRE-SIM can further extend the spatial frequency components of SIM by employing the implicit priors based on the neural network without training datasets. The further super-resolution capability of DRE-SIM is verified by theoretical simulations as well as experimental measurements. Our experimental results show that DRE-SIM can achieve the resolution enhancement by a factor of about 1.4 compared with conventional SIM. Given the advantages of improving the lateral resolution while keeping the imaging speed, DRE-SIM will have a wide range of applications in biomedical imaging, especially when high-speed imaging mechanisms are integrated into the conventional SIM system.