Computational resolution in single molecule localization - impact of noise level and emitter density

Abstract

Classical fluorescence microscopy is a powerful technique to image biological specimen under close-to-native conditions, but light diffraction limits its optical resolution to 200-300 nm-two orders of magnitude worse than the size of biomolecules. Assuming single fluorescent emitters, the final image of the optical system can be described by a convolution with the point spread function (PSF) smearing out details below the size of the PSF. In mathematical terms, fluorescence microscopy produces bandlimited space-continuous images that can be recovered from their spatial samples under the conditions of the classical Shannon-Nyquist theorem. During the past two decades, several single molecule localization techniques have been established and these allow for the determination of molecular positions with sub-pixel accuracy. Without noise, single emitter positions can be recovered precisely - no matter how close they are. We review recent work on the computational resolution limit with a sharp phase transition between two scenarios: 1) where emitters are well-separated with respect to the bandlimit and can be recovered up to the noise level and 2) closely distributed emitters which results in a strong noise amplification in the worst case. We close by discussing additional pitfalls using single molecule localization techniques based on structured illumination.