Detecting the Area of Bovine Cumulus Oocyte Complexes Using Deep Learning and Semantic Segmentation

Abstract

The cumulus-oocyte complex (COC) is an oocyte surrounded by specialized granulosa cells, called cumulus cells. The cumulus cells surrounding the oocyte ensure healthy oocyte and embryo development. The maturity of COCs at oocyte retrieval may be used as an indicator to predict outcome of assisted reproductive technology (ART). Segmenting COCs is a preliminary step in many image processing pipelines to evaluate maturity. However, acquiring well-annotated bright-field microscopy image datasets remains a time-consuming and inaccurate procedure, for most biological domains. Additionally, specialists often partially disagree on their annotations, not only among each other, but also among their own annotations, leading to an inconsistent outcome. Despite the recent advancements in deep learning and image segmentation tools for biological and biomedical images, there is limited usage of them for having more accurate and automated procedures. In this work, we propose an automated pipeline to segment bovine COCs in bright-field microscopy images. The results of our evaluation show that our pipeline is able to segment COCs with the same level of quality as provided by human experts.