Figure 1
From imaging to classification. (a) Widefield polarimetric SHG imaging of the sample at 16 unique polarization state combinations, prepared by the polarization state generator (PSG) and the polarization state analyzer (PSA), which correspond to horizontal linearly polarized (HLP), vertically linearly polarized (VLP), right and left circularly polarized states (RCP and LCP), respectively. (b) Calculation of SHG Stokes vector elements to compute polarimetric parameter images. On the left, image shows breast tissue core pseudo-color images of s0, s1 and s3 component at PSG states of HLP, VLP, RCP and LCP respectively. The right image shows polarimetric parameter maps of the core for SHG intensity, R-Ratio, DCP, SHG-CD and SHG-LD. (c) Subdivision of polarimetric images into 64 sub-images, to allow high-resolution texture analysis and statistical significance testing. (d) Calculation of mean and mean absolute deviation of polarimetric parameter, as well as contrast, correlation, entropy, angular second moment (ASM), and inverse difference moment (IDM) texture parameters of each sub-image. (e) Training of a logistic regression classifier using the polarimetric and texture parameters, and the subsequent prediction to differentiate normal and tumor tissue.
