System- and sample-agnostic isotropic three-dimensional microscopy by weakly physics-informed, domain-shift-resistant...

Area of Science:

• Biomedical Imaging
• Microscopy
• Computational Biology

Background:

• Optical microscopy's diffraction limit restricts axial resolution in 3D imaging.
• Accurate 3D structural analysis is hindered, especially in label-free imaging of thick, heterogeneous tissues.
• Standard imaging assumptions often fail for complex biological samples.

Purpose of the Study:

• To develop a robust framework for isotropic 3D imaging.
• To overcome limitations in axial resolution for label-free imaging of biological tissues.
• To enable accurate 3D structural analysis in challenging samples.

Main Methods:

• Introduced SSAI-3D, a weakly physics-informed, domain-shift-resistant framework.
• Generated a diverse, noise-resilient, sample-informed training dataset.
• Sparsely fine-tuned a large pre-trained blind deblurring network for axial deblurring.

Main Results:

• SSAI-3D demonstrated robust axial deblurring capabilities.
• The framework was successfully applied to label-free nonlinear imaging of organoids, human endometrium, and mouse whisker pads.
• Validated on public datasets of 3D heterogeneous biological tissues with unknown imaging conditions.

Conclusions:

• SSAI-3D provides a robust solution for isotropic 3D imaging.
• The framework enhances axial resolution in challenging label-free microscopy applications.
• Enables more accurate 3D structural analysis of complex biological tissues.