Minimizing In-Vitro Fertilization Failures by Utilizing Artificial Intelligence to Evaluate the Health of Human Embryos

The quality of the transferred embryo is the crucial factor that impacts the success of the in-vitro fertilization (IVF) treatment cycle. Embryologists and other related experts in the field are responsible for deciding which embryo should be chosen to continue in the IVF process. These clinicians are faced with a life-changing task that could potentially lead to failure of conception, and they should ensure their final diagnoses are accurate. Alternative tools used to make this paramount decision are limited, subjective, time-consuming, and extremely expensive. However, an embryologist's skills, coupled with the precision and accuracy of an automated evaluation system, could improve IVF success rates by ensuring consistent results. Employing modern technologies, such as artificial intelligence (AI), is the deciding factor between providing accurate or inaccurate results. This project, LetoHealth, utilizes prevalent convolutional neural network (CNN) architectures to distinguish embryo health and quality at 113 hours post insemination (hpi) on day 5 of culturation based on its morphology. In this study, I assessed ResNET-50, Xception, and custom-built multi-layered CNN architectures; Xception performed the best with a validation accuracy of 0.98, precision of 0.98, and a loss of 0.05 among these. In addition, LetoHealth includes a web-based evaluation tool built using the best-performed Xception model hosted on the cloud that embryologists and clinicians can access, producing instant results of embryo health.