DeepSTORM3D

DeepSTORM3D localizes and reconstructs densely overlapping single-molecule emitters in three-dimensional localization microscopy by combining point-spread-function (PSF) engineering with deep learning to improve axial discrimination and localization accuracy in densely labeled samples.

Key Features:

• PSF engineering: Engineers point-spread functions (PSFs) using additional optical elements to introduce depth-dependent variation in the PSF.
• Depth-varying PSFs: Produces PSFs that vary distinctly with emitter axial position to enable separation of emitters despite lateral PSF overlap.
• Neural network localization: Employs a neural network trained to localize multiple emitters exhibiting densely overlapping Tetrapod PSFs across an extensive axial range.
• PSF design using learning: Uses the trained network to design optimal PSFs tailored for scenarios with multiple emitters.
• Experimental validation: Demonstrated experimental super-resolution reconstructions of mitochondria and volumetric imaging of fluorescently labeled telomeres.
• Improved temporal and spatial performance: Increases temporal resolution and 3D localization accuracy in densely labeled single-molecule localization microscopy data.

Scientific Applications:

• Whole-cell dynamic imaging: Study biological processes in whole cells at timescales that are challenging for conventional localization microscopy.
• Super-resolution reconstruction of mitochondria: Reconstruct mitochondrial structures from dense 3D localization data.
• Volumetric telomere imaging: Perform volumetric imaging of fluorescently labeled telomeres within cells.
• High-temporal-resolution studies: Enable high-temporal-resolution studies of dynamic cellular events in densely labeled samples.

Methodology:

Train a neural network on images of densely overlapping Tetrapod PSFs across an extensive axial range to localize multiple emitters, and use the trained network to design optimal PSFs for multi-emitter conditions.