Y-Net: a one-to-two deep learning framework for digital holographic reconstruction

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Artificial Intelligence

Background:

• Digital holography is a powerful technique for 3D imaging.
• Traditional holographic reconstruction methods can be computationally intensive and may struggle with simultaneous intensity and phase retrieval.
• Deep learning offers potential for accelerating and improving holographic reconstruction.

Purpose of the Study:

• To propose a novel deep learning framework, Y-Net, for digital holographic reconstruction.
• To enable simultaneous reconstruction of intensity and phase information from a single digital hologram.
• To demonstrate the superior performance of Y-Net compared to existing methods.

Main Methods:

• Development of a one-to-two deep learning framework (Y-Net) tailored for holographic reconstruction.
• Training the Y-Net to process digital holograms and output both intensity and phase.
• Comparative analysis of Y-Net's performance against conventional reconstruction techniques.

Main Results:

• The Y-Net successfully reconstructs both intensity and phase information simultaneously from single digital holograms.
• The proposed Y-Net exhibits higher performance with a compact network architecture and reduced parameters.
• Experimental validation using mouse phagocytes demonstrates the practical advantages of the Y-Net.

Conclusions:

• The Y-Net represents a significant advancement in digital holographic reconstruction.
• This deep learning approach offers a more efficient and effective method for quantitative phase imaging.
• The Y-Net shows promise for various applications in biological and materials science imaging.