Advancing Spectrally-Resolved Single Molecule Localization Microscopy using Deep Learning

Spectrally-Resolved Single Molecule Localization Microscopy (srSMLM) is a recent multidimensional technique enriching single molecule localization imaging by the simultaneous recording of single emitters spectra. As for SMLM, the localization precision is fundamentally limited by the number of photons collected per emitters. But srSMLM is more impacted because splitting the emission light from single emitters into a spatial and a spectral channel further reduces the number of photons available for each channel and impairs both spatial and spectral precision - or forces the sacrifice of one or the other. Here, we explored the potential of deep learning to overcome this limitation. We report srUnet - a Unet-based image processing that enhances the spectral and spatial signals and compensates for the signal loss inherent in recording the spectral component. We showed that localization and spectral precision of low-emitting species remain as good as those obtained with a high photons budget together with improving the fraction of localizations whose signal is both spatially and spectrally interpretable. srUnet is able to deal with spectral shift and its application to multicolour imaging in biological sample is straight-forward. srUnet advances spectrally resolved single molecule localization microscopy to achieve performance close to conventional SMLM without complicating its use.