Application of artificial intelligence, deep learning and MLP techniques, to predict live birth in assisted reproduction patients using blastocyst morphology and patient physiological data

Abstract

The assisted reproduction technique - ART is on the rise and is accompanied by constant innovation and modernization. Techniques such as intracytoplasmic sperm injection - ICSI, use of equipment such as time-lapse, preimplantation genetic diagnosis, and screening are currently methods to increase the quality and success of ART. Accompanying this trend, the use of artificial intelligence - AI, in this area, is also being intensively researched either in the determination of gamete quality or selection and in embryo classification. Despite an exponential number of studies, the use of AI within assisted reproduction clinics is not yet a reality, and embryo classification and selection are performed by embryologists who end up incurring intra and inter embryologist errors. In addition, couples who come to clinics have a keen interest in knowing the likelihood of success of ART, but it is very abstract for a human to measure that probably because of the complexity of this process. There are already studies of the use of AI in the prediction of live birth, but these only use the image of the transferred blastocyst, discarding valuable information as physiological aspects of the patient, such as age, BMI, number of oocytes and antral follicles. Thus, the present project aims to implement software using digital processing, artificial neural networks, and genetic algorithms that can predict the probability of pregnancy success based on the morphological data of the blastocyst together with the patient's physiological data. In addition, it is intended a comparative analysis of the use of deep learning and multilayer perceptron - MLP techniques in embryo classification.