Metodologia para criação de mosaico de imagens de lâminas histológicas: uma abordagem de baixo custo usando smartphone e microscópio óptico convencional
Data
2023-02-28
Autores
Santos, Eric Rocha 
Tipo
Dissertação de mestrado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Os ensaios histopatológicos figuram como importante ferramenta para pesquisas
clínicas e diagnósticos. A digitalização das lâminas têm permitido o uso de métodos
computacionais em busca da obtenção de resultados de forma rápida, padronizada e com
maior acurácia. Para tanto, uma alternativa tecnológica que busca automatizar o processo de
aquisição são os scanners. Contudo, seu custo alcança altos patamares, muitas vezes
incompatíveis com a realidade de alguns centros de pesquisas. Desta forma, o objetivo do
presente trabalho consiste em apresentar o desenvolvimento e avaliação do potencial de
descritores de características para a construção do mosaico de imagens a partir de capturas
parciais das lâminas histológicas. A metodologia para o processo de montagem do mosaico é
realizada pela captura de imagens parciais com uma região de sobreposição entre estas,
utilizado um detector de características que sejam capazes de distinguir regiões específicas da
imagem. Essas regiões são descritas através de informações que são extraídas das regiões
detectadas. Em seguida, é verificada a correspondência entre essas regiões descritas para a
construção da matriz de transformação espacial, responsável em definir o deslocamento
necessário para a correta sobreposição das correspondências. Por fim, são aplicados métodos
de Seamless para suavizar as transições entre as tesselas. Para uma sobreposição de 33%, o
ORB combinado com Seamless definida por Dijkstra e Pyramid Blending apresentaram os
melhores resultados. Como conclusão a metodologia proposta se mostrou preciso na
construção do mosaico, desta forma tornando-se uma ferramenta alternativa para centros de
pesquisas e laboratórios que precisam de um processo de baixo custo.
Histopathological assays are an important tool for clinical and diagnostic research. The digitization of the slides has allowed the use of computational methods in search of the management of results in a fast, standardized and more accurate way. Currently, scanners are a technological alternative that seeks to automate the acquisition process. However, its cost reaches high levels, often incompatible with the reality of some research centers. Thus, the objective of this work is to present the development and evaluation of the potential of feature descriptors for the construction of the mosaic of images from partial captures of histological slides. The methodology for the mosaic assembly process is performed by capturing partial images with an overlapping region between them, using a resource detector that is capable of distinguishing specific regions of the image. These regions are described through information that is extracted from the registered regions. Then, the correspondence between these described regions is verified for the construction of the spatial transformation matrix, responsible for defining the necessary displacement for the correct overlapping of the correspondences. Finally, Seamless methods are applied to smooth the transitions between tesserae. For an overlap of 33%, ORB combined with Seamless defined by Dijkstra and Pyramid Blending showed the best results. As a conclusion, the proposed methodology proved to be accurate in the construction of the mosaic, thus becoming an alternative tool for research centers and laboratories that need a low-cost process.
Histopathological assays are an important tool for clinical and diagnostic research. The digitization of the slides has allowed the use of computational methods in search of the management of results in a fast, standardized and more accurate way. Currently, scanners are a technological alternative that seeks to automate the acquisition process. However, its cost reaches high levels, often incompatible with the reality of some research centers. Thus, the objective of this work is to present the development and evaluation of the potential of feature descriptors for the construction of the mosaic of images from partial captures of histological slides. The methodology for the mosaic assembly process is performed by capturing partial images with an overlapping region between them, using a resource detector that is capable of distinguishing specific regions of the image. These regions are described through information that is extracted from the registered regions. Then, the correspondence between these described regions is verified for the construction of the spatial transformation matrix, responsible for defining the necessary displacement for the correct overlapping of the correspondences. Finally, Seamless methods are applied to smooth the transitions between tesserae. For an overlap of 33%, ORB combined with Seamless defined by Dijkstra and Pyramid Blending showed the best results. As a conclusion, the proposed methodology proved to be accurate in the construction of the mosaic, thus becoming an alternative tool for research centers and laboratories that need a low-cost process.