Deep learning-based color holographic microscopy.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Biomedical Engineering

Background:

• Coherent microscopy generates phase and amplitude information but often lacks color.
• Reconstructing full-color images typically requires multiple measurements or complex setups.
• Histopathology relies on color for diagnosis, necessitating accurate color reproduction.

Purpose of the Study:

• To develop a computational framework for high-fidelity color image reconstruction from a single hologram.
• To leverage generative adversarial networks (GANs) for artifact reduction and accurate color transformation.
• To enhance the throughput of coherent microscopy for applications like histopathology.

Main Methods:

• A generative adversarial network (GAN) framework was designed for image reconstruction.
• The network was trained on holographic data illuminated by three different wavelengths.
• The framework processes a single hologram to reconstruct a full-color image.

Main Results:

• The GAN successfully eliminated missing-phase-related artifacts in reconstructed images.
• Accurate color transformation was achieved, yielding high-fidelity color images.
• Experimental validation was performed on stained lung and prostate tissue sections.

Conclusions:

• The proposed GAN-based framework enables high-fidelity color image reconstruction from single-shot holography.
• This method significantly improves the throughput of coherent microscopy systems.
• The framework shows promise for point-of-care histopathology and digital pathology applications.