DeepSTORM3D: dense 3D localization microscopy and PSF design by deep learning.

Area of Science:

• Biophysics
• Microscopy
• Computational Biology

Background:

• Accurate 3D single-molecule localization microscopy (SMLM) is challenging in densely labeled biological samples.
• Point-spread-function (PSF) engineering improves axial localization but struggles with overlapping emitters in dense samples.
• Dense emitter imaging is crucial for enhancing temporal resolution in SMLM.

Purpose of the Study:

• To develop a neural network-based method for localizing multiple, densely overlapping emitters with engineered PSFs in 3D.
• To design an optimal PSF for multi-emitter localization in dense SMLM.
• To enable high-speed, high-resolution 3D imaging of cellular structures.

Main Methods:

• Training a neural network to localize multiple emitters with overlapping Tetrapod PSFs over a broad axial range.
• Utilizing the trained network to computationally design an optimal PSF for dense emitter scenarios.
• Experimental validation using super-resolution imaging of mitochondria and volumetric imaging of telomeres.

Main Results:

• Successfully trained a neural network (DeepSTORM3D) for accurate multi-emitter localization with overlapping Tetrapod PSFs.
• Demonstrated experimental application for high-resolution imaging of cellular components.
• Achieved volumetric imaging of fluorescently labeled telomeres in cells.

Conclusions:

• DeepSTORM3D overcomes the challenge of localizing dense, overlapping emitters in 3D SMLM.
• The developed approach enables the study of biological processes in whole cells at unprecedented timescales.
• This method significantly advances the capabilities of localization microscopy for dynamic cellular studies.