Computational multi-wavelength phase synthesis using convolutional neural networks [Invited].

Area of Science:

• Optical microscopy
• Biophysics
• Computational imaging

Background:

• Multi-wavelength digital holographic microscopy (MWDHM) measures refractive indices indirectly.
• Existing MWDHM methods are unsuitable for dynamic samples or require complex optical setups.

Purpose of the Study:

• To develop a deep learning-based computational phase synthesis method for high-speed, single-shot MWDHM.
• To enable accurate refractive index estimation without increasing experimental complexity.

Main Methods:

• Convolutional neural networks (CNNs) were used for computational phase synthesis.
• Data-driven techniques were employed for hologram synthesis, phase synthesis, and phase-to-index prediction.
• Simulations of holographic recording and cell phantoms were used for validation.

Main Results:

• Accurate dual-wavelength hologram and phase synthesis were achieved.
• Direct phase-to-index prediction from single-wavelength holograms was demonstrated.
• Effective 2D phase unwrapping with discontinuities was performed.

Conclusions:

• Deep learning offers a novel computational approach for high-speed MWDHM.
• This method allows for single-shot refractive index estimation without complex optics.
• The validated computational concept shows promise for studying dynamic biological samples.