Predição do sucesso de gestação utilizando algoritmos de machine learning após procedimentos de fertilização in vitro realizados por um serviço de atendimento público de saúde.

• Ano de defesa: 2021
• Autor(a) principal: Nayara Cristina Nunes Barreto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; FARMACIA - FACULDADE DE FARMACIA; Programa de Pós-Graduação em Análises Clínicas e Toxicológicas; UFMG
• Palavras-chave em Português: Infertilidade; FIV / ICSI; Modelo de predição; Machine learning; Random forest
• Link de acesso: http://hdl.handle.net/1843/49514

Infertility, defined as the absence of conception after an period of 12 months of unprotected sexual intercourse, has become a global health problem. Infertility affects about 37 to 70 million couples worldwide, which justifies the increase in the number of couples looking for in vitro fertilization (IVF). Despite advances and technical improvements in artificial fertilization procedures, some couples are unable to succeed due to the high complexity of the technique and several variables, controllable or not, that can compromise the final result. Machine Learning (ML) is a sub-area of Artificial Intelligence that is dedicated to the study of algorithms and statistical models to perform a procedure without the need for the use of explicit instructions, in order to generate predictive models of the outcome. The objective of this study was to apply ML models to predict the success of pregnancy after the IVF procedure in a public health service, including pre-implantation variables. This study included the analysis of a database comprising 771 cases of couples undergoing IVF at the Hospital das Clínicas of the Federal University of Minas Gerais, between 2013 and 2019. The ML-based algorithms were used: Logistic Regression, Randon Forest, XG Boost and Support Vector Machines (SVM). The Random Forest algorithm showed the best performance, with better accuracy, sensitivity and area under the ROC curve, indicating the 20 variables, among the 90 totals in the database, that are most relevant to predict the success of pregnancy after IVF. Our ML algorithm can be useful for predicting pregnancy in women undergoing IVF, as well as defining the variables in which clinical intervention can improve treatment success.