Methods to improve the understanding of microscopic embryo data sets using image analysis

In this work we propose a framework for visualisation of semi-automatically segmented microscopic images of human embryos. A large part of the education of biologists consists of learning to interpret the output from a variety of analytical methods and medical imaging modalities, which can be more or less abstract in nature. Even in visual microscopy, the optical setup and the different ways to increase contrast between the sample and background produce image artefacts which have to be taken into account when interpreting the image. In in vitro fertilisation, the correct evaluation of the quality of the embryo is crucial for successful future development of the implanted foetus. Embryos are selected for transfer based on a number of characteristics, such as blastomere symmetry, degree of fragmentation and number and size of blastomeres. Traditionally, this evaluation has to a large extent been done by manual observation through visual microscopy, and obtaining the necessary expertise takes years of training. Here we show how the output from different analytical methods may be combined and how creative visualisation and improved user interaction with large data sets may improve the understanding of the sample under study. We show how existing computer-aided tools can be used in embryo selection and discuss automation as a way to quantify the subjective bias of manual embryo selection. We use data from human embryos as a case study, but the methods may be applied to any type of biological or microscopic material.