Virtual-scanning light-field microscopy for robust snapshot high-resolution volumetric imaging.

Area of Science:

• Biomedical Imaging
• Microscopy
• Computational Biology

Background:

• High-speed 3D intravital imaging is crucial for studying dynamic subcellular processes in biological systems.
• Light-field microscopy (LFM) offers snapshot 3D imaging with low phototoxicity but suffers from limited resolution and reconstruction artifacts.

Purpose of the Study:

• To develop a novel deep learning framework, virtual-scanning LFM (VsLFM), to overcome the resolution limitations of conventional LFM.
• To achieve diffraction-limited resolution in snapshot 3D LFM for enhanced intravital imaging.

Main Methods:

• A physics-based deep learning framework, VsLFM, was developed to computationally increase LFM resolution.
• A large dataset (40 GB) of high-resolution scanning LFM data across species was constructed.
• Physical priors between phase-correlated angular views were exploited to address frequency aliasing and bypass hardware scanning.

Main Results:

• VsLFM successfully increased LFM resolution up to the diffraction limit within a single snapshot.
• The method enabled ultrafast 3D imaging at speeds up to 500 volumes per second.
• VsLFM demonstrated effective imaging of diverse biological processes, including embryonic zebrafish heartbeats, Drosophila brain activity, and mouse liver neutrophil migration.

Conclusions:

• VsLFM offers a significant advancement in high-speed, high-resolution 3D intravital microscopy.
• The framework overcomes LFM's inherent resolution and artifact limitations, enabling detailed visualization of rapid biological events.
• VsLFM has broad applicability for studying dynamic cellular and subcellular functions in various living organisms.