Inteligência artificial na predição de risco de ventilação mecânica em casos graves de Covid-19

• Ano de defesa: 2023
• Autor(a) principal: Godoy, Mariana Frizzo de
• Orientador(a): Padoin, Alexandre Vontobel
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Medicina e Ciências da Saúde
• Departamento: Escola de Medicina
• País: Brasil
• Palavras-chave em Português: COVID-19; Tomografia Computadorizada; Inteligência Artificial
• Palavras-chave em Inglês: COVID-19; Computed Tomography; Artificial Intelligence
• Área do conhecimento CNPq: CIENCIAS DA SAUDE::MEDICINA
• Link de acesso: https://tede2.pucrs.br/tede2/handle/tede/10753

Introduction: In December 2019, the city of Wuhan, China, became the center of an outbreak of pneumonia of unknown cause, which would soon have the new coronavirus, SARS-CoV-2, as the identified cause of the disease. In March 2020, WHO declared the infection a pandemic situation. In the first year of the pandemic, it was extremely important to quickly identify patients with the greatest potential for poor outcome and need for mechanical ventilation, especially for the organization of intensive care unit beds, which were often scarce in Brazil. Objective: The present study aims to determine the accuracy of chest computed tomography, evaluated by a machine learning model, in predicting the need for mechanical ventilation in patients hospitalized with SARS due to COVID-19. Methods: This is a retrospective cohort study, carried out in two Brazilian hospitals. CT scans of patients hospitalized for severe acute respiratory syndrome and COVID-19 confirmed by RT-PCR were included. An artificial intelligence model based on neural convolution networks was developed. Results: The sensitivity of the model was 0.417 and the specificity 0.860. The corresponding area under the ROC curve for the test set was 0.68. Conclusion: A machine processing model with high specificity was created, capable of reliably predicting patients who will not need mechanical ventilation.