Deep Learning Enabled Holographic Polarization Microscopy

Polarization microscopy has long been used in various fields due to its unique capability of highlighting birefringent objects. Traditional polarization microscopy techniques usually require the collection of two or more images from light paths with different polarization states to either enhance the image contrast or retrieve quantitative information of birefringent specimen. Because of this, these methods typically have complex optical designs and require experienced technicians to operate. Here, we present a deep learning-based holographic polarization microscopy framework which transforms the holographic amplitude and phase information of a sample into the birefringent retardance and orientation channels. This framework only requires the addition of one polarizer/analyzer pair to an existing lensfree holographic imaging system, with a compact optical design and a large field of view (~20-30 mm²). We experimentally tested this framework with different types of birefringent samples including monosodium urate (MSU) crystals, showing its capability to accurately reconstruct quantitative birefringence information of specimen.