Mueller matrix microscopy for digital pathology

A Mueller matrix image (MMI) contains rich optical and microstructural information in all the pixels, which are comprehensively encoded in the high-dimensional polarization feature space. Since the contrast mechanisms of MMI do not rely on absorption and are sensitive to scattering features down to a fraction of wavelength, the technique is label-free and non-invasive, capable of cross-scale and quantitative measurements and ideal for probing biomedical samples including tissues and cells. Despite all the apparent advantages of Mueller matrix microscopy for quantitative characterization of the microstructural properties, how to extract polarization features of pathological significance remains challenging for clinical applications. In this report, we present polarization super-pixels approach to characterize approximately the complex pixel distributions of MMIs in polarization space and use machine learning approaches to exploit polarization features for diagnosis applications in clinical pathology.