Combining deep learning with SUPPOSe and compressed sensing for SNR-enhanced localization of overlapping emitters

BRINATTI VAZQUEZ, GUILLERMO D.; MARTINEZ, OSCAR E.; LACAPMESURE, AXEL M.; MARTINEZ, SANDRA; MAZZEO, ALEJANDRO

Applied Optics

The Optical Society

Lugar: Washington; Año: 2022 vol. 61 p. 39 - 49

1559-128X

We presentgSUPPOSe, a novel, to the bestof our knowledge, gradient-based implementation of the SUPPOSealgorithm that we have developed for the localization of singleemitters. We study the performance of gSUPPOSe and compressed sensingSTORM (CS-STORM) onsimulations of single-molecule localization microscopy (SMLM) imagesat different fluorophore densities and in a wide range ofsignal-to-noise ratio conditions. We also study the combination ofthese methods with prior image denoising by means of a deepconvolutional network. Our results show that gSUPPOSe can address thelocalization of multiple overlapping emitters even at a low number ofacquired photons, outperforming CS-STORM in our quantitative analysisand having better computational times. We also demonstrate that imagedenoising greatly improves CS-STORM, showing the potential of deeplearning enhanced localization on existing SMLM algorithms. Thesoftware developed in this work is available as open source Pythonlibraries.