Proposta e avaliação de um método de segmentação para Alberta Stroke Program Early CT Score

• Ano de defesa: 2021
• Autor(a) principal: Brito, Rafael de Freitas
• 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 Elétrica
• Palavras-chave em Português: Acidente Vascular Cerebral; Alberta Stroke Program Early CT Score; Diagnóstico Assitido; Processamento Digital de Imagens; Registro de Atlas; Tomografia Computadorizada; Stroke,; Assisted Diagnosis; Digital Image Processing; Atlas Recording; Tomography computerized; CNPQ::ENGENHARIAS; Engenharia elétrica
• Link de acesso: https://repositorio.ufu.br/handle/123456789/32492; http://doi.org/10.14393/ufu.di.2021.5524

Stroke is one of the leading causes of death and impairment in the world, affecting specially the population of low and middle income countries. The Alberta Stroke Program Early CT Score (ASPECTS) is a score that quantifies the extent of early ischemic changes in stroke victims' tomographies, being widely used as a prognostics and patient selection tool. ASPECTS, however, lacks reliability (low interrater agreement), such that methods for assisted and even automated diagnostics might improve its reliability, and therefore stroke care. Many automated ASPECTS evaluation tools have already been developed, both in an academic and comercial context, and although all of them utilize some sort of region segmentation method, segmentation is often mentioned very briefly or not even discussed in ASPECTS' studies. Given that gap in literature, this study aims to present and evaluate an ASPECTS region segmentation algorithm, with public libraries and databases. The algorithm here presented is divided in four steps: pre-processing, global registration, local registration and visualization/evaluation, and was evaluated in both quantitative and qualitative terms. For the quantitative evaluation, the method obtained a mean Dice coeficient of 0.6587 with a 0.0595 standard deviation and a mean Hausdorff distance of 14.3903 with a standard deviation of 4.4366 for all 10 computed tomographies evaluated. The qualitative evaluation was done through the subjective evaluation of a neurologist (0 to 10), and resulted in a mean score of 8.44 with a standard deviation of 0.726. Although the method presented satisfying results, it can still be improved, using post-processing techniques, for exemple. It's also important that the method is tested in a larger number of tomographies, to improve its robustness.