Deep learning facilitated whole live cell fast super-resolution imaging

A fully convolutional encoder-decoder network (FCEDN), a deep learning model, was developed and applied to image scanning microscopy (ISM). Super-resolution imaging was achieved with a 78 mu m x 78 mu m field of view and 12.5 Hz-40 Hz imaging frequency. Mono and dual-color continuous super-resolution images of microtubules and cargo in cells were obtained by ISM. The signal-to-noise ratio of the obtained images was improved from 3.94 to 22.81 and the positioning accuracy of cargoes was enhanced by FCEDN from 15.83 +/- 2.79 nm to 2.83 +/- 0.83 nm. As a general image enhancement method, FCEDN can be applied to various types of microscopy systems. Application with conventional spinning disk confocal microscopy was demonstrated and significantly improved images were obtained.