Segmentação dos ossos do joelho a partir de imagens de tomografia computadorizada

• Ano de defesa: 2020
• Autor(a) principal: Waltz, Flavio da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Biomédica; UTFPR
• Palavras-chave em Português: Hounsfield, Godfrey, 1919-; Tomografia; Joelhos; Ossos; Reconstrução de imagens; Imagem tridimensional; Processamento de imagens - Técnicas digitais; Tomography; Knee; Bones; Image reconstruction; Three-dimensional imaging; Image processing - Digital techniques; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA::BIOENGENHARIA; Engenharia Biomédica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/5274

The segmentation of anatomical structures based on medical images, such as computed tomography, has been applied to improve practices such as surgical planning, prosthesis manufacturing, simulations and more. In the orthopedic area, 3d bone models can be used, for example, in the manufacture of personalized guides and implants. Precisely extracting complex elements from an image is not a simple task. Most related research proposes methods for segmenting a single anatomical structure. In this research, it was decided to test the difficulties related to the segmentation of the surface of the four bones that constitute the knee joint (fêmur, patella, tibia and fibula) using computed tomography images and the Hounsfield units. The proposed method consists of a processing sequence that begins with the resampling of the data, which are then pre-segmented by global thresholding with a chosen threshold based on the value proposed by the literature for bone segmentation: 200 HU. Then, the discontinuities of the obtained meshes, when present, are corrected through morphological closure, and undesirable objects, when present, are removed by morphological erosion. Finally, the models are segmented by Active Contour Without Edges executed by morphological operators. Four case studies with four different individuals were carried out using this method. The results obtained show that the proposed method is capable of segmenting and generating 3d models of the four knee bones, allowing each one to be manipulated individually if necessary. However, the bone segmentation based on Hounsfield units did not generate the expected result in three of the four case studies, of which only one had its bones correctly segmented with the application of a threshold equal to 200 HU during pre-segmentation by global thresholding. In all other cases, the threshold had to be lowered só that the bone mesh was segmented as expected.