Application of Artificial Intelligence - Multi-Input Convolutional Neural Networks - for Predicting Gestational Success in Assisted Reproduction

• Grant number: 25/10242-9
• Support Opportunities: Scholarships in Brazil - Master
• Start date: October 01, 2025
• End date: September 30, 2027
• Field of knowledge: Health Sciences - Medicine
• Principal Investigator: José Celso Rocha
• Grantee: Bruno Araújo Mendes
• Host Institution: Faculdade de Ciências e Letras (FCL-ASSIS). Universidade Estadual Paulista (UNESP). Campus de Assis. Assis , SP, Brazil
• Associated research grant: 23/16156-1 - Prediction of gestational success, using the variables of morphology and morphokinetics of the human embryo and the patient, using the Artificial Intelligence technique, through deep learning and multilayer perceptron., AP.R

Abstract

Assisted reproduction has evolved significantly since Lazzaro Spallanzani's pioneering experiments in the 18th century to modern techniques such as in vitro fertilization and intracytoplasmic sperm injection. In recent decades, assisted reproduction has undergone important technological advances, with the digitization of laboratory processes and the increasing use of embryo imaging and analysis systems. This modernization has favored the development of more objective and standardized methods for embryo evaluation, with the potential to reduce the subjectivity of traditional analyses and increase the accuracy in selecting embryos with the highest implantation potential. Despite these advances, embryo selection still depends on subjective morphological evaluations, which can limit the effectiveness of treatments. The use of incubators with time-lapse technology has enabled continuous, non-invasive monitoring of embryonic development, paving the way for the application of artificial intelligence in the analysis of these images. Among the most promising techniques are convolutional neural networks (CNNs), which are capable of identifying complex visual patterns associated with embryonic viability. This project proposes the development of an architecture based on multi-input CNNs for the simultaneous analysis of segmented images of three key structures: internal cell mass, trophectoderm, and blastocoel. Each structure will be processed by a specific network, and their outputs will be combined into a unified model. In addition, a fourth input will be incorporated into the system, based on a Multilayer Perceptron network, responsible for processing 33 quantitative morphological variables digitally extracted from the segmented images. These variables include measures such as area, perimeter, circularity, and textural attributes, offering a complementary and structured view of the embryos. By integrating visual and numerical information, the hybrid model aims to increase predictive accuracy, reduce subjectivity, and support more assertive clinical decisions. This is expected to contribute to the standardization of embryo selection and improve success rates in assisted reproduction.