Chapter Twelve - Superresolution Image Processing

Abstract

Optical diffraction places a limit on the resolution of the optical microscope. Superresolution microscopy is a collection of techniques for imaging with resolution beyond the diffraction limit. Deconvolution can improve resolution up to the diffraction limit but is limited by noise in the image. Statistical methods fill in some of the missing spectral components with their most likely values, to extrapolate spatial frequencies beyond the band limit and produce the most likely image. Stimulated emission depletion microscopy (SEDM) achieves resolution beyond the diffraction limit by spatially deactivating fluorophores outside of a small central illuminated point. Single-molecule localization microscopy (SMLM) captures single emitting fluorophores one at a time and determines their precise position by fitting a Gaussian function. Structured illumination microscopy (SIM) works by illuminating the sample with nonuniform patterns of light and combining multiple images in a postprocessing step. Machine learning methods train a neural network to learn common relationships between the available spatial frequencies and the missing ones, enabling the algorithm to “guess” the amplitude of the missing components.