Selection of Single Potential Embryo to Improve the Success Rate of Implantation in IVF Procedure using Machine Learning Techniques

Abstract

Classification and grading of human In-Vitro Fertilized (IVF) embryos is time consuming and challenging. Various factors like morphological and genetic quality fertilized egg, its sensitivity to environmental factors like temperature, and pH, etc make the decision making process hard. In absence of an embryo selection metric, embryologists will not be able to estimate the success of implantation apriori. Therefore developing a statistically provable automaton could help embryologists increase success of IVF procedure. Such a method helps the embryologists to overcome the intra and inter observer variations in estimating the quality of embryo. In this paper, several Machine Learning (ML) methods have been tested to develop a model that helps selection of a single potential embryo with high success rate. Fertilized eggs are observed over a period extending up to five days. Time-lapse photos are used to train the model. A candidate elimination algorithm builds a version space using a test set. Then a work space is built to test the input data. Cloud based GPU services have also been tested. They can achieve almost 85% accuracy as compared to a manual (visual) validation procedure. Algorithmic methods gave an yield of 78.4%, which is acceptable in many cases to reduce work load. Most of these models are practically useful to predict the implantation rate and outcome of IVF.