Prediction of fetal heartbeat through artificial intelligence and morphological, morphokinetic and patient-related variables

Author(s)

Dóris Spinosa Chéles

• Document type: Master's Dissertation
• Press: Botucatu. 2022-09-29.
• Institution: Universidade Estadual Paulista (Unesp). Instituto de Biociências. Botucatu
• Defense date: 2022-03-02
• Advisor: Marcelo Fábio Gouveia Nogueira

Abstract

Assisted reproduction technologies have been increasingly used and improved over time. For embryo transfer, choosing the best quality embryo is a decisive step for the pregnancy occurrence. For this, embryologists have classification systems at the moment of embryos evaluation. However, this evaluation depends on the embryologists’ experience and, consequently, is influenced by intrinsic subjectivity, both between different evaluators and by the same evaluator. In addition, knowledge about the times of embryo development - facilitated by the implementation of the time-lapse system in the laboratory routine - and about the characteristics related to the patient undergoing assisted reproduction treatment are factors that can help at the time of embryo selection, and potentially improve the pregnancy rate achieved in clinics. Therefore, an objective evaluation is necessary to improve the success of these technologies. Thus, three sets of distinct input variables (blastocyst morphology, embryo morphokinetics and patient-related characteristics) were used for application in artificial intelligence (artificial neural networks and genetic algorithm). For morphology, digital image processing of human blastocysts was performed to obtain predictive variables of blastocyst quality. For morphokinetics, parameters related to the times of embryo development were used. For the patient, scores were developed for each input variable based on a literature search. These sets of variables were trained and simulated separately and together to predict fetal heartbeat. In addition, the three sets together were also used to predict live birth. Considering the three sets together as input, the best artificial neural network found for the prediction of fetal heartbeat had an overall accuracy of 95.2% in training and 78.4% in simulation. The best artificial neural network for live birth prediction had an overall accuracy of 98.0% in training and 81.1% in simulation. Thus, the application of artificial intelligence has the potential to objectively assist embryologists during the choice of the most suitable embryo for transfer, and thus, moving from the in silico (laboratory) field towards clinical practice in assisted reproduction. (AU)

• FAPESP's process: 20/07634-9 - Prediction of fetal heartbeat through artificial intelligence and morphological, morphokinetic and patient-related variables
• Grantee: Dóris Spinosa Chéles
• Support Opportunities: Scholarships in Brazil - Master