High-throughput, high-resolution Generated Adversarial Network Microscopy
share
We for the first time combine generated adversarial network (GAN) with wide-field light microscopy to achieve deep learning super-resolution under a large field of view (FOV). By appropriately adopting prior microscopy data in an adversarial training, the network can recover a high-resolution, accurate image of new specimen from its single low-resolution measurement. This capacity has been adequately demonstrated by imaging various types of samples, such as USAF resolution target, human pathological slides and fluorescence-labelled fibroblast cells. Their gigapixel, multi-color reconstructions verify a successful GAN-based single image super-resolution procedure. Furthermore, this deep learning-based imaging approach doesn;t necessarily introduce any change to the setup of a conventional wide-filed microscope, reconstructing large FOV (about 95 mm^2), high-resolution (about 1.7 μm) image at a high speed (in 1 second). As a result, GAN-microscopy opens a new way to computationally overcome the general challenge of high-throughput, high-resolution microscopy that is originally coupled to the physical limitation of system's optics.
READ FULL TEXT