Journal of Nanophotonics 2020

SuperSTED

Deep Adversarial Network for Super Stimulated Emission Depletion Imaging

Mengzhou Li, Hongming Shan, Sergey Pryshchep, Maria M. Lopez, Ge Wang

Current or former lab member. Select a name to see all of their projects.

Method overview of SuperSTED
The deep neural network structure (Fig. 2 of the paper).

Abstract

STED microscopy can reach a lateral resolution of about 20 nm, but in live-cell imaging it typically reaches about 100 nm because of phototoxicity, leaving many structures unresolved. This work proposes a deep adversarial network that refines a STED image to a higher resolution. Training pairs are simulated from physical models of the point spread function and of cell structures.

The network achieved 30-nm resolution from 60-nm STED images, with structural similarity index values of about 0.98, higher than Lucy–Richardson deconvolution and a UNet-based network.

What’s in the repository

  • A notebook for network training and one for testing
  • The data are described in the README: shuffled training patch pairs, plus validation and test sets

Citation

@article{li2020deep,
  title   = {Deep Adversarial Network for Super Stimulated Emission Depletion Imaging},
  author  = {Li, Mengzhou and Shan, Hongming and Pryshchep, Sergey and Lopez, Maria M. and Wang, Ge},
  journal = {Journal of Nanophotonics},
  volume  = {14},
  number  = {1},
  pages   = {016009},
  year    = {2020},
  doi     = {10.1117/1.JNP.14.016009}
}
Back to projects