SNSMIL, a real-time single molecule identification and localization algorithm for super-resolution fluorescence microscopy

Abstract

Single molecule localization based super-resolution fluorescence microscopy offers significantly higher spatial resolution than predicted by Abbe’s resolution limit for far field optical microscopy. Such super-resolution images are reconstructed from wide-field or total internal reflection single molecule fluorescence recordings. Discrimination between emission of single fluorescent molecules and background noise fluctuations remains a great challenge in current data analysis. Here we present a real-time, and robust single molecule identification and localization algorithm, SNSMIL (Shot Noise based Single Molecule Identification and Localization). This algorithm is based on the intrinsic nature of noise, i.e., its Poisson or shot noise characteristics and a new identification criterion, QSNSMIL, is defined. SNSMIL improves the identification accuracy of single fluorescent molecules in experimental or simulated datasets with high and inhomogeneous background. The implementation of SNSMIL relies on a graphics processing unit (GPU), making real-time analysis feasible as shown for real experimental and simulated datasets.

Automatic Bayesian single molecule identification for localization microscopy

“…SNSMIL 8 is a SMLM method that makes use of a single user-dependent characteristic, Q SNSMIL . The threshold of Q SNSMIL was determined using the Auto-Bayes method.…”

“…, we showed here that the selection characteristic SNR wavelet obtained in the wavelet segmentation algorithm 9 10 is suitable for the Auto-Bayes method. On the other hand, for the selection characteristic Q SNSMIL used in SNSMIL 8 there are cases where it is more difficult or maybe even impossible to confidently distinguish false and true emitters. In any case a suitable distribution model needs to be determined for each chosen characteristic.…”

“…The pixel values of the source image are converted to photoelectron counts using Equation 1 8 . Here n ( i , j ) is the photoelectron count of pixel ( i , j ); I ( i , j ) is the value of pixel ( i , j ); I offset denotes the bias offset; G eff is the applied effective electronic gain in AD counts per photoelectron (note for an EMCCD camera this value is dependent on the applied electron multiplication gain).…”

“…In all packages, it is necessary to define characteristics of a true emitter in order to distinguish single emitter signals from background fluctuations. A variety of characteristics have been used such as signal-to-noise ratio (SNR) 8 9 10 11 12 , spot brightness 12 13 14 and point spread function (PSF) size 11 13 . Depending on the software, these characteristics can be calculated using a variety of methods, like wavelet filtering 9 10 , Gaussian filtering 8 10 or median filtering 10 15 .…”

The Directional Observation of Highly Dynamic Membrane Tubule Formation Induced by Engulfed Liposomes

“…This way, object patterns could be resolved which are smaller than the diffraction limit. Finally, a well-resolved dSTORM image was achieved by fitting a Gaussian distribution function to the intensity of the observed fluorescence spots (obtained via the SNSMIL method 44 ) from thousands of frames. All analyses were carried out using a custom-written MATLAB code.…”