Utilização de inteligência artificial no suporte ao diagnóstico a partir de dados hospitalares da COVID-19 e de suas sequelas pós-aguda

• Ano de defesa: 2024
• Autor(a) principal: Farrapo, Ruann Campos de Castro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Detecção COVID-19; Sequelas da COVID-19; Extração de características; Predição; Machine learning; Inteligência artificial; COVID-19 detection; COVID-19 sequels; Feature extraction; Prediction; CNPQ::ENGENHARIAS
• Link de acesso: http://repositorio.ufc.br/handle/riufc/79052

COVID-19 infections and their Post-Acute Sequelae of COVID-19 (SPAC) represent a global health crisis. Associated with this, COVID-19 has had a profound impact on the health of people around the world. In addition to the direct consequences of virus infection, such as serious illness and death, there has been a significant increase in levels of stress, anxiety and depression due to fear of the disease, social isolation and uncertainty about the future. Futhermore, understanding the risks associated with COVID-19, its sequelae and its biological mechanisms has not yet been fully established. Given this gap, it is crucial to develop an extractive and predictive approach to support identifying both COVID-19 and its possible sequelae. Therefore, the present study proposes a methodology to carry out this detection and prediction using Artificial Intelligence (AI) techniques related to Machine Learning (ML). This approach involves the utilization of several classifiers, including Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), neural network Multilayer Perceptron (MLP), K — Nearest Neighbors (KNN) and Light Gradient Boosting Machine (LGBM). In addition to their individual constructions, these classifiers were combined, forming a new ensemble classifier. These models are applied to two different databases. The first refers to the detection of COVID-19, containing 400 positive records and 691 negative records, with 16 variables. The second set of data is aimed at SPACs, covering examinations of patients with different conditions: 174 with Hypertension, 181 with Asthma, 182 with Congestive Heart Failure and 190 with Coronary Artery Disease. The results obtained highlight the effectiveness of the proposed approach, with an accuracy result of 97% for the first database and an average accuracy of 88,75% for the second database. These accuracy results demonstrate the model’s ability to predict both the presence of COVID-19 and its possible sequelae, providing a valuable tool to support clinical practice and public health.