Emprego de Inteligência Artificial na predição de desfechos gestacionais

Abstract

The health of women of reproductive age is considered a critical factor by the Pan American Health Organization (PAHO/WHO). The Health Sciences Center at the Federal University of Recôncavo da Bahia (UFRB) is conducting research on health management and planning, fo cusing on this population. The goal is to use computational tools to improve care for women and implement an efficient care model. The study is being conducted in family health units and involves women of childbearing age and pregnant women aged between 15 and 49 years. A Research Project was created that uses computational intelligence to improve maternal and child health indicators. Care during pregnancy involves several variables and their multiple associations in controlling problems that may occur. In this work, a model for predicting birth outcomes, such as normal or premature birth, is being developed, using artificial intelligence (AI) and plasma levels from pregnant women's laboratory tests as input attributes. The use of AI and Machine Learning (ML) is becoming increasingly common in the field of pregnancy, offering solutions for monitoring and predicting outcomes. There are different AM methods and structures available, as demonstrated in the reviewed papers. However, the complexity and quantity of health variables, together with the need to protect data confidentiality, make it dif ficult to determine the best path for AI to follow in the search for relationships between varia bles and prediction of outcomes. of pregnancy. This area still requires a lot of research, both to identify the best AI method/structure and to establish these relationships. This work, in addition to developing the proposed model for the Research Project, also verifies and compares the pre dictions of gestational outcomes between some AI techniques and the created model. This opens up the possibility for further research to verify the validity of the relationships found and ex plore those that were not established during model development. The methodology used was Fuzzy Logic, which is rarely found in the reviews carried out, mainly in relation to the objective proposed in this study. The inclusion of this AI technique contributes to filling the existing gap in the available options for the health area. The results showed a level of accuracy of predictions of outcomes around 70%, which is compatible with other verified algorithms. One of the ad vantages of Fuzzy Logic is that it does not require a large dataset for training, as most AI tech niques do, as it does not need this step. Based on the knowledge of specialists in the field, parameters for input attributes and output classes are established. Few instances of data are needed to fit and evaluate the final selection of output classes, which is important in situations with scarce data