Accelerating multicolor spectroscopic single-molecule localization microscopy using deep learning.

Area of Science:

• Biophysics
• Microscopy
• Computational Imaging

Background:

• Spectroscopic single-molecule localization microscopy (sSMLM) enables multicolor super-resolution imaging at the nanoscale.
• Current sSMLM techniques require extensive frame acquisition, limiting imaging speed.
• Spectral cross-talk in multicolor sSMLM complicates accurate imaging of multiple fluorophores.

Purpose of the Study:

• To develop a computational method for accelerating multicolor sSMLM imaging.
• To overcome limitations of long acquisition times and spectral cross-talk in sSMLM.
• To enable high-density multicolor super-resolution imaging with reduced data requirements.

Main Methods:

• Utilized deep convolution neural networks (CNNs) for image reconstruction.
• Trained CNNs on sSMLM datasets to reconstruct images from fewer frames.
• Developed a strategy to handle spectral cross-talk and improve image fidelity.

Main Results:

• Reconstructed high-density multicolor super-resolution images using up to 8-fold fewer frames.
• Achieved significant reduction in imaging acquisition time without compromising spatial resolution.
• Demonstrated successful two-color and three-color imaging of cellular structures (tubulin, mitochondria, peroxisomes) in fixed cells.

Conclusions:

• The presented computational strategy effectively accelerates multicolor sSMLM imaging.
• Deep learning enables faster super-resolution microscopy with reduced data acquisition.
• This technique enhances the practicality of sSMLM for multicolor biological imaging.