Abstract:The advance in microscopy and artificial intelligence enables the application of digital pathology in various classification situations to help pathologists reduce the challenge of performing diagnosis purely based on their visualization experience. Human endometrium is receptive to the embryo only during a defined period in a menstrual cycle. The endometrial phase characterization is crucial for the formation of a healthy pregnancy. Polarization imaging is an emerging label-free and non-invasive technique that is good at characterizing the microstructures of biological tissues. In this study, polarization imaging was combined with digital pathology to characterize the microstructures of endometrium samples at the typical proliferative phase and typical secretory phase. The involved polarization parameters include Muller matrix polar decomposition (MMPD) derived parameters δ, θ and a set of rotation invariant parameters PL, DL, qL, rL and their corresponding angular parameters αP, αD, αq and αr. The approaches for the digitalization of the polarization parameter images include the statistical mean analysis that does not involve image texture information, the Local Binary Pattern (LBP) analysis that involves partial image texture information, and the machine learning classifications that make full use of the polarization parameter image information. A class distance Score was defined to evaluate the performance of polarization parameters in the statistical mean and the image texture analysis. The statistical mean analysis indicates parameter DL that relate to the dichroism of the endometrial tissues shows the best class separation ability with the highest class distance Score. Image texture analysis indicates parameter DL still has the highest class distance Score. And compared with the statistical mean method, the class distance Score for DL increased after LBP process. The results of machine learning classification show parameter αD classified by Convolutional Neural Network (CNN) architecture 1 and parameter αP classified by CNN architecture 2 have the same highest accuracy of 87%. This study shows the potential of applying the digital pathology techniques on polarization parameter images to achieve endometrial phase characterization.

Oocytes Polar Body Detection for Automatic Enucleation

A Real-Time Oocyte Polar Body Detection Method

Image Classification of Chicken Embryo Based on Matched Filter and Skeleton Curvature Feature

Automated Ovarian Cancer Identification Using End-to-End Deep Learning and Second Harmonic Generation Imaging

Automatic Cell Rotation Based on Real-Time Detection and Tracking

Oocyte biology and genetics revelations from polar bodies

Segmentation of mature human oocytes provides interpretable and improved blastocyst outcome predictions by a machine learning model

Stain-free detection of embryo polarization using deep learning

Automatic evaluation of human oocyte developmental potential from microscopy images

Grading of Mammalian Cumulus Oocyte Complexes using Machine Learning for in Vitro Embryo Culture

Polar Bodies in Assisted Reproductive Technology: Current Progress and Future Perspectives1

Stain-free enucleation of mouse and human oocytes with a 1033 nm femtosecond laser

Automated Denudation of Oocytes

OoCount: A machine-learning based approach to mouse ovarian follicle counting and classification

Automated Detection of Polycystic Ovary Syndrome from Ultrasound Images

Improved accuracy for estrous cycle staging using supervised object detection

Visual Servoed Robotic Mouse Oocyte Rotation

Functional Human Oocytes Generated by Transfer of Polar Body Genomes

Polarization Imaging Feature Characterization of Different Endometrium Phases by Machine Learning

Automatic stage identification of Drosophila egg chamber based on DAPI images

Genome Analyses of Single Human Oocytes