Análise da ploidia de embriões humanos por meio da Inteligência Artificial com o uso de variáveis de morfologia, morfocinética e variáveis relacionadas com a paciente
Carregando...
Arquivos
Data
2022-06-24
Autores
Souza, Rebeca Colauto Milanezi de
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A infertilidade é considerada um problema de saúde mundial que afeta cerca de 15% dos casais em idade reprodutiva. As atuais tecnologias empregadas pelas técnicas de reprodução assistida não são totalmente eficazes, sendo a taxa de sucesso gestacional consideravelmente baixa (cerca de 25 a 40%). Assim, a escolha do embrião de melhor qualidade é fundamental para o sucesso das duas principais técnicas de reprodução assistida, a fertilização in vitro e a injeção intracitoplasmática de espermatozoides. A taxa de sucesso dessas técnicas pode ser influenciada pela presença de aneuploidias embrionárias, as quais podem resultar em falha de implantação, em abortos e mesmo em malformações nos nascidos vivos. Para evitar técnicas invasivas de análise genética dos embriões, (por exemplo, a biópsia embrionária que utiliza irreversivelmente parte das células embrionárias para a análise genética), a Inteligência Artificial associada ao processamento de imagens do blastocisto, pode vir a ser um caminho alternativo. Nesse sentido, essa técnica pode vir a viabilizar a análise genética embrionária de forma não invasiva, a um baixo custo e de forma objetiva. Com isso, o objetivo deste trabalho foi adaptar um software já desenvolvido (como uma plataforma inicial) para a previsão da ploidia dos embriões humanos, distinguindo os embriões euploides dos embriões aneuploides. Para isso foram utilizadas as técnicas de Inteligência Artificial, como as Redes Neurais Artificiais e os Algoritmos Genéticos, associadas com a aplicação das variáveis oriundas do processamento digital de imagens dos blastocistos (variáveis morfológicas), da morfocinética (tempos de clivagem e de desenvolvimento do embrião) e variáveis da paciente em conjunto com as variáveis celulares que possam auxiliar na verificação da ploidia embrionária. No total, 574 embriões – com respectivos dados da morfocinética e dados das pacientes – foram avaliados. Esses dados foram obtidos de pacientes submetidas a ciclos de fertilização de maio de 2019 a março de 2021. A melhor rede neural artificial obtida foi aquela em que foram aplicadas as variáveis da morfologia, morfocinética e da paciente com as variáveis celulares (com diagnóstico específico para os fatores de infertilidade). Nesta rede, o acerto geral foi de 98,6% no treinamento, de 80,0% no teste, de 69,2% na simulação e de 64,7% no teste-cego. Com esses resultados, é possível concluir que a rede neural artificial obtida tem uma acurácia considerável na predição da classificação embrionária quanto à ploidia, sendo uma alternativa para viabilizar a análise genética embrionária de forma não invasiva, a um baixo custo e de forma objetiva. No entanto, apesar dos resultados promissores, concluímos que essas tecnologias precisam ser analisadas em estudos prospectivos e não somente em estudos retrospectivos, como o do presente trabalho, objetivando uma melhor análise de sua eficácia.
Infertility is considered a worldwide health problem that affects about 15% of couples of reproductive age. The current technologies used by assisted reproduction techniques are not fully effective, because pregnancy success rates are considerably low (about 25 to 40%). Thus, choosing the best quality embryo is essential for the success of the two main assisted reproduction techniques, in vitro fertilization and intracytoplasmic sperm injection. The success rates of these techniques can be influenced by the presence of embryonic aneuploidies, which can result in implantation failure, miscarriages or malformations in live births. To avoid invasive techniques for genetic analysis of embryos (for example, embryo biopsy that irreversibly uses part of the embryonic cells for genetic analysis), Artificial Intelligence associated with blastocyst image processing may be an alternative way. In this sense, this technique may enable embryo genetic analysis in a non-invasive way, at a low cost and in an objective way. Therefore, the objective of this work was to adapt an already developed software (as an initial platform) for the prediction of ploidy of human embryos, distinguishing euploid embryos from aneuploid embryos. For this, Artificial Intelligence techniques were used, such as Artificial Neural Networks and Genetic Algorithms, associated with the application of variables arising from digital processing of blastocyst images (morphological variables), morphokinetics (cleavage and embryo development times) and patient variables together with cellular variables that could help verify embryonic ploidy. In total, 574 embryos – with respective morphokinetic and patient datas – were evaluated. These data were obtained from patients undergoing fertilization cycles from May 2019 to March 2021. The best artificial neural network obtained was the one in which the morphology, morphokinetic and patient variables were applied with cellular variables (with specific diagnosis for infertility factors). In this network, the overall accuracy was 98.6% in the training, 80.0% in the test, 69.2% in the simulation and 64.7% in the blind test. With these results, it is possible to conclude that artificial neural network obtained has considerable accuracy in predicting the embryonic classification in terms of ploidy, being an alternative to enable the embryonic genetic analysis in a non-invasive way, at a low cost and in an objective. However, despite the promising results, we conclude that these technologies need to be analyzed in prospective studies and not only in retrospective studies, such as the one in the present work, aiming at a better analysis of their effectiveness.
Infertility is considered a worldwide health problem that affects about 15% of couples of reproductive age. The current technologies used by assisted reproduction techniques are not fully effective, because pregnancy success rates are considerably low (about 25 to 40%). Thus, choosing the best quality embryo is essential for the success of the two main assisted reproduction techniques, in vitro fertilization and intracytoplasmic sperm injection. The success rates of these techniques can be influenced by the presence of embryonic aneuploidies, which can result in implantation failure, miscarriages or malformations in live births. To avoid invasive techniques for genetic analysis of embryos (for example, embryo biopsy that irreversibly uses part of the embryonic cells for genetic analysis), Artificial Intelligence associated with blastocyst image processing may be an alternative way. In this sense, this technique may enable embryo genetic analysis in a non-invasive way, at a low cost and in an objective way. Therefore, the objective of this work was to adapt an already developed software (as an initial platform) for the prediction of ploidy of human embryos, distinguishing euploid embryos from aneuploid embryos. For this, Artificial Intelligence techniques were used, such as Artificial Neural Networks and Genetic Algorithms, associated with the application of variables arising from digital processing of blastocyst images (morphological variables), morphokinetics (cleavage and embryo development times) and patient variables together with cellular variables that could help verify embryonic ploidy. In total, 574 embryos – with respective morphokinetic and patient datas – were evaluated. These data were obtained from patients undergoing fertilization cycles from May 2019 to March 2021. The best artificial neural network obtained was the one in which the morphology, morphokinetic and patient variables were applied with cellular variables (with specific diagnosis for infertility factors). In this network, the overall accuracy was 98.6% in the training, 80.0% in the test, 69.2% in the simulation and 64.7% in the blind test. With these results, it is possible to conclude that artificial neural network obtained has considerable accuracy in predicting the embryonic classification in terms of ploidy, being an alternative to enable the embryonic genetic analysis in a non-invasive way, at a low cost and in an objective. However, despite the promising results, we conclude that these technologies need to be analyzed in prospective studies and not only in retrospective studies, such as the one in the present work, aiming at a better analysis of their effectiveness.
Descrição
Palavras-chave
Inteligência artificial, Ploidia, Embrião, Reprodução assistida, Morfologia, Morfocinética, Artificial intelligence, Ploidy, Embryo, Assisted reproduction, Morphology, Morphokinetics