HYPER-Net: Physics-Conditioned Self-Supervised Reconstruction for Fourier Light-Field Microscopy.

Area of Science:

• Biomedical Imaging
• Computational Microscopy
• Machine Intelligence in Biology

Background:

• High-speed, high-resolution 3D imaging is crucial for understanding dynamic biological systems.
• Existing methods face challenges with large datasets, computational demands, and optical variations.

Purpose of the Study:

• To introduce HYPER-Net, a physics-conditioned self-supervised framework for Fourier light-field microscopy.
• To enable fast, robust 3D reconstruction for volumetric imaging.

Main Methods:

• Developed HYPER-Net, integrating scan-free acquisition with 3D reconstruction.
• Incorporated experiment-specific point-spread functions to enforce measurement consistency and modulate features.
• Utilized a self-supervised approach reducing reliance on ground-truth data.

Main Results:

• Demonstrated high-fidelity volumetric imaging of diverse biological samples including organoids, embryos, and C. elegans.
• Achieved accurate imaging of tissue morphology, cardiac function, and neural/muscular dynamics.
• Showcased robustness to system variations and generalizability across contexts.

Conclusions:

• HYPER-Net offers a versatile framework for rapid volumetric imaging and quantitative analysis.
• The approach enhances accessibility and applicability of advanced microscopy techniques.
• Facilitates breakthroughs in basic research and biomedical applications for dynamic biological systems.