AUTHOR=Shen Lei , Zhang Yanran , Chen Wenfeng , Yin Xinghui 

TITLE=The Application of Artificial Intelligence in Predicting Embryo Transfer Outcome of Recurrent Implantation Failure

JOURNAL=Frontiers in Physiology

VOLUME=13

YEAR=2022

URL=https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2022.885661

DOI=10.3389/fphys.2022.885661

ISSN=1664-042X

ABSTRACT=<p><bold>Background:</bold> Recurrent implantation failure (RIF) refers to that infertile patients have undergone multiple <italic>in vitro</italic> fertilization (IVF) or intracytoplasmic sperm injection (ICSI) cycles and transferred multiple embryos without embryo implantation or clinical pregnancy. Due to the lack of clear evidence-based medical guidelines for the number of embryos to be transferred in RIF patients, how to obtain the highest single cycle pregnancy success rate with as few embryos transferred as possible while avoiding multiple pregnancy as much as possible, that is, how to balance the pregnancy success rate and multiple pregnancy rate, is a great challenge for clinicians and RIF patients. We urgently need an effective and reliable assisted decision-making method to help clinicians find this balance, and an artificial intelligence (AI) system will provide an efficient solution.</p><p><bold>Design and Methods:</bold> In this research, we filtered out the RIF data set (<italic>n</italic> = 45,921) from the Human Fertilisation and Embryology Authority (HFEA) database from 2005 to 2016. The data set was divided into two groups according to the number of embryos transferred, Group A and B. Group A included 34,175 cycles with two embryos transferred, while Group B included 11,746 cycles with only one embryo transferred, each containing 44 features and a prediction label (pregnancy). Four machine learning algorithms (RF, GBDT, AdaBoost, and MLP) were used to train Group A and Group B data set respectively and 10-folder cross validation method was used to validate the models.</p><p><bold>Results:</bold> The results revealed that the AdaBoost model of Group A obtained the best performance, while the GBDT model in Group B was proved to be the best model. Both models had been proved to provide accurate prediction of transfer outcome.</p><p><bold>Conclusion:</bold> Our research provided a new approach for targeted and personalized treatment of RIF patients to help them achieve efficient and reliable pregnancy. And an AI-assisted decision-making system will be designed to help clinicians and RIF patients develop personalized transfer strategies, which not only guarantees efficient and reliable pregnancy, but also avoids the risk of multiple pregnancy as much as possible.</p>