Video-Rate Quantitative Phase Imaging Using a Digital Holographic Microscope and a Generative Adversarial Network.

Area of Science:

• Biomedical Optics
• Microscopy
• Computational Imaging

Background:

• Conventional off-axis digital holographic microscopy (DHM) requires extensive computational processing for accurate phase reconstruction.
• Standard methods involve spatial filtering and tilt compensation, often leading to long processing times that limit real-time imaging of dynamic biological processes.
• Numerical focusing may also be necessary for distortion-free quantitative phase images.

Purpose of the Study:

• To develop a fast and robust quantitative phase imaging method for DHM using artificial intelligence.
• To overcome the limitations of conventional computational reconstruction techniques in DHM.
• To enhance the visualization of biological specimens through improved phase imaging.

Main Methods:

• Implementation of a conditional generative adversarial network (cGAN) for DHM image reconstruction.
• Training and validation of the cGAN model using human red blood cells.
• Utilizing an off-axis Mach-Zehnder DHM system for data acquisition.

Main Results:

• The cGAN model provides stable background levels in reconstructed phase images, improving specimen visualization.
• The learning-based method achieves computationally efficient reconstruction.
• DHM images were reconstructed seven times faster compared to conventional computational approaches.

Conclusions:

• The proposed cGAN offers a significant speed improvement for DHM quantitative phase imaging.
• This AI-driven approach enhances the robustness and visualization capabilities of DHM for biological samples.
• The method shows promise for enabling video-rate imaging of dynamic biological processes using DHM.