Avaliação numérica da dose ocupacional em procedimentos intervencionistas utilizando métodos de deep learning em tempo real

• Ano de defesa: 2024
• Autor(a) principal: Tavares, Otavio Joaquim
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Biomédica
• Palavras-chave em Português: Dose de órgãos; Organ Dose; Previsão espacial.; Spatial Prediction; Exposição médica; Medical Exposition; Modelos de imagem; Imaging Models; Monitoramento durante procedimentos; Procedure Monitoring; Inteligência Artificial; Artificial Intelligence; CNPQ::CIENCIAS EXATAS E DA TERRA; Engenharia biomédica; ODS::ODS 4. Educação de qualidade - Assegurar a educação inclusiva, e equitativa e de qualidade, e promover oportunidades de aprendizagem ao longo da vida para todos.
• Link de acesso: https://repositorio.ufu.br/handle/123456789/43633; http://doi.org/10.14393/ufu.di.2024.124

This study addresses the assessment of dose to organs/tissues, based on the spatial position of the physician during the interventional radiology procedure. The main objective was to analyze the prediction capacity, considering different distances in relation to the X-ray tube, in different organs/tissues. The methods applied included a prediction model with low training cost, however, with specific limitations, intended for a type of procedure considering immobile patients, fixed gender and static conditions of the X-ray equipment. Restricting the physician’s movement within the marker area was crucial to reducing image deformation. The results highlighted that, in small organs, such as the prostate, the greatest differences occurred between the predicted dose values and those calculated with MCNP6.2. The results highlight the feasibility of integrating the Multilayer Perceptron (MLP) model, used to predict doses in each organ, with the YOLOv4 model, for real-time identification of the doctor’s spatial positions.This integration has the advantage of being low cost, thus being a crucial resource for monitoring the dose values of the medical team during interventional radiology procedures, although there are still opportunities for improvement identified for both models.