Undergrad Thesis: Classification of Human Embryo Viability using Deep Learning Techniques.
In Vitro Fertilization (IVF) is a common procedure for treating infertility across the world. IVF involves externally inseminating a human egg and transferring the resulting fertilized egg (embryo) into the patient’s uterus. Clinicians rely on visual inspection of morphological features of an embryo to determine its viability. Such inspections make the selection procedure subjective and susceptible to the clinician’s bias and error. According to the Canadian Fertility and Andrology Society’s 2017 report, in 2016 the overall percentage of live births using IVF was at 39.5% for patients under the age of 35. Unfortunately, IVF treatment is expensive and emotionally challenging, which makes more than one treatment hard for many couples. Correctly choosing viable embryos is important for improving the outcome of IVF treatments. Deep learning techniques have found great success in medical research tasks and have been utilized for human embryo segmentation and cell counting. We propose a deep learning method for classifying the viability of a human embryo using Attention Networks. The network incorporates a main branch and a mask branch. The main branch is used to extract features and the mask branch regularizes these features to draw attention to the most important morphological features of the embryos.