Deep Learning-Based Holographic Polarization Microscopy.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Pathology Diagnostics

Background:

• Polarized light microscopy (PLM) is crucial for high-contrast imaging of birefringent specimens in pathology.
• Traditional PLM systems are complex, costly, and require expert operation due to multiple optical paths.
• There is a need for simplified, cost-effective PLM techniques for broader diagnostic applications.

Purpose of the Study:

• To develop a deep learning-based holographic microscope for quantitative birefringence analysis.
• To simplify polarized light microscopy by integrating it with a lensfree holographic imaging system.
• To enable accurate retardance and orientation measurements from a single polarization state.

Main Methods:

• A deep neural network was trained to process phase-recovered holograms from a lensfree holographic system.
• The system requires only a polarizer/analyzer pair added to the inline holographic setup.
• The neural network reconstructs images equivalent to single-shot computational polarized light microscopy (SCPLM).

Main Results:

• The deep learning model successfully extracted quantitative birefringence retardance and orientation information.
• The method utilizes sample morphology and holographic amplitude/phase data for birefringence extraction.
• Imaging of monosodium urate and triamcinolone acetonide crystals demonstrated comparable results to SCPLM.

Conclusions:

• The deep learning-based holographic microscope offers a simplified and cost-effective alternative to traditional PLM.
• The system achieves high accuracy in quantitative birefringence measurements.
• Its simpler design and larger field-of-view hold potential for expanding PLM accessibility, especially in resource-limited settings.