DEFESA DE TESE DE DOUTORADO Nº 13

Aluno: Maicon Herverton Lino Ferreira da Silva Barros

Título: “Machine-learning for classification of the prognosis of tuberculosis using real data from Brazil”

Orientadora: Patricia Takako Endo - (PPGEC)

Coorientador: Vanderson Souza Sampaio - (UFPE)

Examinadora Externa: Gisely Cardoso de Melo (UFAM)

Examinadora Externa: Danielle Moura - (UPE/FENSG)

Examinador Interno: Dimas Cassimiro - (PPGEC)

Examinador Interno: Cleyton Rodrigues - (PPGEC)

Data-hora: 21 de fevereiro de 2024, às 14:00h.
Local:Formato Remoto e Presencial.


Resumo:

         Tuberculosis (TB), for many years until the advent of COVID-19, was the leading cause of death by an infectious agent worldwide. Despite efforts by the World Health Organization (WHO) to reduce the incidence of tuberculosis, it is estimated that in 2021, about 10.6 million people fell ill with the disease and 1.6 million deaths were recorded globally. In Brazil, one person contracts tuberculosis every five minutes and one dies every hour from the disease; in 2020 alone, there was an increase of 12% in the number of deaths compared to 2019. Monitoring the possible outcomes of a patient with tuberculosis is an important task that can help reduce early mortality in a patient diagnosed with this disease. However, determining this outcome is not a trivial task, especially in terms of anticipating the patient’s prognosis. For decades, the state of health and quality of life during the treatment of a disease have been receiving increasing attention in the health field. Brazil has the Information System for Notifiable Diseases (SINAN), which contains a database with records of patients with compulsory notification diseases, including tuberculosis. Classifying the outcome of tuberculosis treatment into categories of cure and death (prognosis) using a tool that employs a machine learning model, can assist health professionals in making decisions about the most appropriate treatment, given the individual conditions of the patient and the likely course of the disease based on medical experience. Depending on the prognosis, late or inadequate treatment can result in unsatisfactory outcomes, including exacerbation of clinical symptoms, poor quality of life, and increased risk of death. In this thesis, we propose the development of a tool named TITO that uses artificial intelligence (AI) to assist in the prognosis of tuberculosis. The tool has four modules: decision support; application; interface; ongoing monitoring. Regarding the decision support module, preprocessing of SINAN data from 2001 to April 2020 was carried out, with about 1.7 million patients containing clinical, laboratory, and sociodemographic data of patients who were treated for pulmonary and extrapulmonary tuberculosis, in addition to the application of machine learning techniques, feature selection, and random search to find hyperparameter optimization. Through a rigorous scientific methodology, experiments were conducted with different scenarios of data balancing and imbalancing, using appropriate metrics to evaluate the models with the objective of selecting the artificial intelligence model with the best performance for TITO. The application module consists of a computer program, developed in Python, that runs on a web application server. The program uses the trained AI model to perform classifications of tuberculosis prognosis; it performs interpretability through the XAI technique and controls data entry for the follow-up module. The interface module, responsible for data entry, is responsible for sending the information received from the user to the application module and also for retrieving the classification history from the follow-up module. The follow-up module for monitoring the prognosis over time is responsible for recording and monitoring the prognosis and probabilities of cure or death to maintain a history of predictions over time. Promising results were found with the use of machine learning models. In the usability evaluation, TITO achieved an 83.95% score. Finally, TITO will allow health vii professionals to monitor patients’ prognosis classifications, making decision-making supported by a tool using artificial intelligence.

Go to top Menú