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

Area of Science:

• Biophysics
• Optical Microscopy
• Computational Biology

Background:

• Single-molecule localization microscopy (SMLM) enables super-resolution imaging by localizing individual fluorescent molecules.
• Accurate localization of dense and overlapping emitters remains a challenge in SMLM.
• Existing algorithms like compressed sensing STORM (CS-STORM) have limitations in low photon count and high density scenarios.

Purpose of the Study:

• To introduce gSUPPOSe, a novel gradient-based implementation of the SUPPOSe algorithm for single-emitter localization.
• To evaluate the performance of gSUPPOSe against CS-STORM under various conditions, including different fluorophore densities and signal-to-noise ratios.
• To investigate the impact of deep learning-based image denoising on SMLM localization algorithms.

Main Methods:

• Development of gSUPPOSe, a gradient-based algorithm for single-emitter localization.
• Simulations of SMLM images with varying fluorophore densities and signal-to-noise ratios.
• Application of a deep convolutional network for image denoising prior to localization.
• Quantitative performance analysis and comparison of gSUPPOSe and CS-STORM.

Main Results:

• gSUPPOSe effectively localizes multiple overlapping emitters even at low photon counts.
• gSUPPOSe outperforms CS-STORM in quantitative analysis and demonstrates superior computational efficiency.
• Deep learning-based image denoising significantly enhances the performance of CS-STORM.
• The study highlights the synergistic potential of deep learning with existing SMLM algorithms.

Conclusions:

• gSUPPOSe offers an advanced solution for single-emitter localization in SMLM, particularly for challenging dense and low-photon conditions.
• Deep learning integration presents a promising strategy to improve the performance of current SMLM localization techniques.
• The developed open-source software facilitates further research and application in super-resolution microscopy.