Avaliação da segmentação de microcalcificações em mamografias com operações morfológicas

• Autor(a) principal: Goes, Bruno Augusto Machado
• Data de Publicação: 2024
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/43293; http://doi.org/10.14393/ufu.di.2024.635
• topic: Câncer de mama; Mamografia; Microcalcificação; Segmentação; Watershed; Breast cancer; Mammography; Microcalcification; Segmentation; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA; Engenharia biomédica; Mamografia; Calcificação; Mamas - Câncer; ODS::ODS 3. Saúde e bem-estar - Assegurar uma vida saudável e promover o bem-estar para todos, em todas as idades.

Mammography is an x-ray imaging method recommended for breast cancer early detection, the second most diagnosed cancer type worldwide, because allows to visualize suspicious findings, such as masses and microcalcifications, in early stages of disease. Microcalcifications are small calcium deposits that could be associate to breast cancer, being important the detection of these structures to correct image evaluation. In this scenario, research in digital mammography image processing aims to develop methods to detect and segmentate microcalcifications to facilitate the searching for these objects in image. This work intents to perform microcalcification segmentation using watershed segmentation with markers defined by the morphological operators h-maxima and top-hat. To achieve this, tests were carried out with different structuring element size, minimum intensity difference between pixels, thresholding and denoising techniques, such as the Wiener filter and wavelet transform. Besides that, two artificial neural networks were developed for classification of true and false positive regions, aiming to reduce the number of false positive regions. The marker-controlled watershed segmentation using markers from h-maxima and top-hat operations was well succeeded in identifying regions with microcalcifications, resulting in high sensitivity value, but with low specificity and accuracy values, due to high false positive numbers. The results of the classification step indicate that the developed models were effective to reduce the number of false positive regions, with increase in accuracy and specificity, but decrease in sensitivity.