A carregar...
Publicação
Classificação automática da expressão HER2em imagens de hibridização in situ de tecido mamário
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 10.69 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A sobre-expressão do gene Human epidermal growth factor receptor (HER2) é um factor associado a pior prognóstico em tumores da mama. Avaliar a expressão deste gene no tumor torna-se um aspecto importante para as decisões dos procedimentos terapêuticos a seguir. O status HER2 do tumor pode ser determinado com base em testes de anatomia patológica, que recorrem a técnicas de hibridização in situ. Este tipo de teste implica a contagem de pontos de hibridização correspondentes ao gene HER2 e ao centrómero do cromossoma 17 (CEP17) e de núcleos. Este teste implica um procedimento moroso e repetitivo por parte dos médicos patologistas. A proposta deste estudo foi analisar e construir uma solução, no contexto da patologia digital, que permitisse automatizar os processos de contagem necessários na avaliação da expressão HER2 em tumores da mama. Desenvolveu-se um sistema, baseado em Python, que recorre apenas a ferramentas open-source. Aplicaram-se métodos computacionais e estatísticos de modo a ler as imagens, segmentar os núcleos e sinais de HER e CEP17, extrair features, classificar e proceder a contagens. A segmentação dos pontos de hibridização foi feita através da deteção de blobs e a sua classificação com base num modelo de regressão logística. Segmentação Watershed e deteção de blobs foram analisados como hipóteses de segmentação dos núcleos. Decidiu-se usar a deteção de blobs também como método para contagem de núcleos. Estabeleceu-se uma comparação entre os resultados do algoritmo e contagens manuais que permitiu tirar algumas conclusões. Métodos de deteção de blobs são alternativas válidas para as contagens requeridas no teste do HER2. Existiram erros que afetam na mesma medida as contagens automáticas de HER2 e CEP17. A contagem de núcleos não é independente das contagens dos pontos de hibridização. Mesmo quando as contagens diferiram, nos casos analisados, os parâmetros utilizados para classificar a expressão HER2 são coincidentes entre contagens automáticas e manuais. Como trabalho futuro, o algoritmo desenvolvido pode ser incluído num software open source de análise e processamento de imagens de anatomia patológica. Com base em processos semelhantes aos implementados, é possível extrair mais dados com potencial de abrir novos horizontes no que diz respeito à caracterização da expressão HER2 de tumores.
Overexpression of the gene human epidermal growth factor receptor (HER2) is a factor associated with a prognosis in breast tumors. Evaluating a gene for this tumor is an important aspect to define the therapeutic procedures to follow. The HER2 status of the tumor can be defined based on pathological anatomy tests, which use in situ hybridization techniques. This type of assay involves the counting of hybridization points corresponding to the HER2 gene and the chromosome 17 centromere (CEP17) and nuclei. The purpose of this study was to analyze and develop a digital pathology solution that allows the automation of processes in the evaluation of HER2 in breast tumors. A system, based on Python, was developed using only open source tools. Computational and statistical methods were applied to images to segment HER and CEP17 nuclei and signals, extract features, classify and count. The segmentation of the hybridization points was performed through the detection of blobs and their classification based on a logistic regression model. Watershed segmentation and blob detection were considered hypotheses to segment nuclei. It was decided to use a blob detector to find and count the nuclei. A comparison was established between the results of the algorithm and the manual counts. Use blob detectors is a viable approach to perform the required counts in HER2 test. There were errors that affected the measure of HER2 and CEP17 in the same way. The nuclei count is not independent of the hybridization point counts. Even when the counts differed, the cases were, the parameters used to classify the HER2 expression are coincident between automatic and manual counts. As future work, the developed algorithm can be included in an open source software for analysis and processing of pathological anatomy images. Based on similar procedures with the implements, it is possible to extract more data with the potential to open new horizons with respect to the characterization of HER2 expression of tumors.
Overexpression of the gene human epidermal growth factor receptor (HER2) is a factor associated with a prognosis in breast tumors. Evaluating a gene for this tumor is an important aspect to define the therapeutic procedures to follow. The HER2 status of the tumor can be defined based on pathological anatomy tests, which use in situ hybridization techniques. This type of assay involves the counting of hybridization points corresponding to the HER2 gene and the chromosome 17 centromere (CEP17) and nuclei. The purpose of this study was to analyze and develop a digital pathology solution that allows the automation of processes in the evaluation of HER2 in breast tumors. A system, based on Python, was developed using only open source tools. Computational and statistical methods were applied to images to segment HER and CEP17 nuclei and signals, extract features, classify and count. The segmentation of the hybridization points was performed through the detection of blobs and their classification based on a logistic regression model. Watershed segmentation and blob detection were considered hypotheses to segment nuclei. It was decided to use a blob detector to find and count the nuclei. A comparison was established between the results of the algorithm and the manual counts. Use blob detectors is a viable approach to perform the required counts in HER2 test. There were errors that affected the measure of HER2 and CEP17 in the same way. The nuclei count is not independent of the hybridization point counts. Even when the counts differed, the cases were, the parameters used to classify the HER2 expression are coincident between automatic and manual counts. As future work, the developed algorithm can be included in an open source software for analysis and processing of pathological anatomy images. Based on similar procedures with the implements, it is possible to extract more data with the potential to open new horizons with respect to the characterization of HER2 expression of tumors.
Descrição
Tese de mestrado integrado, Engenharia Biomédica e Biofísica (Sinais e Imagens Médicas) Universidade de Lisboa, Faculdade de Ciências, 2018
Palavras-chave
Patologia Digital Processamento de Imagem Expressão HER2 Teses de mestrado - 2018