Large-scale, single-shot quantitative polarization imaging with deep learning for label-free brain-wide analysis

We introduce PolarGAN, a physics-aware generative model for achieving single-shot quantitative polarization imaging (QPLI). Built upon a customised patch-based network, polarGAN retrieve polarization parameters (PPs) from single bright-field image, bypassing the need for multiple images under different illumination states. This approach rapidly generates brain-wide PPs mappings with high-resolution (1.35 μm), large-scale (2 cm x1.4 cm), reducing imaging time and hardware complexity compared to traditional methods, showing the potentials for accurate QPLI of dynamic samples. We also developed a PPs-based strategy for automated brain regional birefringence analysis, providing new insights into high-resolution birefringence heterogeneity of brain structure.