Implementação e avaliação de metodologia para construção de mosaico aplicado a imagens de lâminas histológicas de tumor neural
Data
2023-12-09
Autores
Santos, Eric Rocha 
Tipo
Trabalho de conclusão de curso
Título da Revista
ISSN da Revista
Título de Volume
Resumo
A crescente demanda por diagnósticos
histopatológicos, vem destacando desafios como a escassez de
patologistas e a demora no processo de identificação de
doenças. A digitalização de amostras histológicas é crucial
para o desenvolvimento de técnicas de Machine Learning. No
entanto, o alto custo da digitalização em países em
desenvolvimento requer alternativas de baixo custo. O
trabalho propõe o uso de técnicas de Image Stitching (IS) para
a construção de mosaicos em imagens histopatológicas de
tumores neurais, visando uma abordagem mais acessível.
Foram aplicados algoritmos de Detecção e Descrição de
Características, Correspondência de Características e
Transformação Espacial para criar mosaicos a partir de
imagens parciais. As técnicas de Seamless foram utilizadas
para minimizar distorções. Os resultados indicam que a
metodologia proposta preserva eficientemente a informação
nas Regiões de Interesse (ROI). O uso de algoritmos como
SIFT e SURF, juntamente com técnicas de Seamless,
demonstrou alta correlação com a imagem original, com o
algoritmo SURF destacando-se pela eficiência computacional.
O estudo destaca a importância de buscar alternativas de
digitalização acessíveis, essenciais para a democratização do
diagnóstico histopatológico em países com recursos limitados.
A metodologia proposta apresenta uma solução viável,
contribuindo para a eficiência do processo diagnóstico.
The growing demand for histopathological diagnoses has brought to light challenges such as a shortage of pathologists and delays in the disease identification process. The digitization of histological samples is crucial for the development of Machine Learning techniques. However, the high cost of digitization in developing countries calls for low-cost alternatives. This work proposes the use of Image Stitching (IS) techniques for constructing mosaics in histopathological images of neural tumors, aiming for a more accessible approach. Algorithms for Feature Detection and Description, Feature Matching, and Spatial Transformation were applied to create mosaics from partial images. Seamless techniques were employed to minimize distortions. The results indicate that the proposed methodology efficiently preserves information in Regions of Interest (ROI). The use of algorithms such as SIFT and SURF, along with Seamless techniques, showed a high correlation with the original image, with the SURF algorithm standing out for computational efficiency. The study emphasizes the importance of seeking affordable digitization alternatives, essential for democratizing histopathological diagnosis in resource-limited countries. The proposed methodology offers a viable solution, contributing to the efficiency of the diagnostic process
The growing demand for histopathological diagnoses has brought to light challenges such as a shortage of pathologists and delays in the disease identification process. The digitization of histological samples is crucial for the development of Machine Learning techniques. However, the high cost of digitization in developing countries calls for low-cost alternatives. This work proposes the use of Image Stitching (IS) techniques for constructing mosaics in histopathological images of neural tumors, aiming for a more accessible approach. Algorithms for Feature Detection and Description, Feature Matching, and Spatial Transformation were applied to create mosaics from partial images. Seamless techniques were employed to minimize distortions. The results indicate that the proposed methodology efficiently preserves information in Regions of Interest (ROI). The use of algorithms such as SIFT and SURF, along with Seamless techniques, showed a high correlation with the original image, with the SURF algorithm standing out for computational efficiency. The study emphasizes the importance of seeking affordable digitization alternatives, essential for democratizing histopathological diagnosis in resource-limited countries. The proposed methodology offers a viable solution, contributing to the efficiency of the diagnostic process