Noise-free quantitative phase imaging in Gabor holography with conditional generative adversarial network

Area of Science:

• Optical Imaging
• Biomedical Optics
• Computational Imaging

Background:

• Gabor holography is susceptible to superimposed twin-image noise, degrading phase image quality.
• Quantitative phase imaging is crucial for analyzing biological samples.
• Existing noise reduction methods in holography can be complex or limited.

Purpose of the Study:

• To develop a deep learning approach for eliminating twin-image noise in Gabor holography.
• To quantitatively assess the model's performance in recovering noise-free phase images.
• To explore the model's ability to generalize to unseen biological structures and compensate for experimental imperfections.

Main Methods:

• A conditional generative adversarial network (C-GAN) was trained using synthetic Gabor holograms and noise-free off-axis digital holography phase images.
• Gabor holograms were synthesized by manipulating frequency-domain images and adding DC terms.
• Fresnel approximation was used for digital propagation to generate phase images for model input.

Main Results:

• The C-GAN model successfully eliminated superimposed twin-image noise from Gabor holographic phase images.
• Quantitative analysis showed high similarity between model-recovered and actual noise-free phase images.
• The model demonstrated an ability to reconstruct previously unseen elliptical cell lines and compensate for reconstruction distance errors.

Conclusions:

• Deep learning, specifically C-GAN, offers a robust solution for noise reduction in Gabor holography.
• The developed model shows potential for accurate quantitative phase imaging of biological samples.
• The model's generalization capabilities and error compensation highlight its practical utility in optical microscopy.