Assinaturas de entropia amostral para classificação de imagens H&E de linfomas não-Hodgkin: uma análise envolvendo técnicas de segmentação e normalização de cores
Data
2020-09-29
Autores
Rozendo, Guilherme Botazzo
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Neste trabalho é apresentado um método baseado em assinaturas de texturas da entropia amostral para quantificar e classificar os grupos linfoide crônica, linfoma folicular e linfoma de células do manto do câncer linfoma não-Hodgkin. O poder discriminativo das assinaturas de entropia amostral foi testado via coloração com hematoxilina e eosina, e diferentes técnicas de segmentação e normalização de cores. As assinaturas de textura foram definidas como curvas de características obtidas a partir de múltiplas observações, combinando os parâmetros m e r da entropia amostral. O comportamento de cada assinatura foi determinado a partir das métricas área sob a curva, obliquidade, ponto máximo de entropia e razão de área. Os resultados foram características fornecidas para classificadores baseados em função, lazy learning, árvores, evolução genética, enxame de partículas e comportamento animal. A melhor combinação de características e classificador foi testada com a adição de diferentes níveis de ruído. Os desempenhos conquistados com a proposta foram taxas de acurácia entre 98,72% e 99,60% com apenas 10 atributos, o que superou os resultados fornecidos por importantes técnicas disponíveis na literatura, mostrando que a abordagem é promissora e pode ser útil para o estudo e reconhecimento de padrões do câncer linfomas não-Hodgkin.
Herein, we present a method that uses entropy texture signatures to distinguish chronic lymphocytic leukemia, follicular lymphoma and mantle cell lymphoma from non-Hodgkin’s lymphoma. The discriminative power of sample entropy signatures were tested via hematoxylin and eosin staining, and multiple segmentation techniques and color normalization approaches. Texture signatures were defined as feature curves obtained from multiple observations, by associating m and r parameters. The behavior of each signature was determined from areas under the curve, skewness, maximum entropy values and area ratios. The results consisted of features which were provided for different functions, such as lazy learning, trees, genetic evolution, particle swarm and animal behavior-based classifiers. The tightest associations between features and classifiers were tested as different noise levels were added. The performance achieved with the proposed method was accuracy rates between 98.72% and 99.60% with only 10 attributes, that is greater than results provided by previously established techniques, which indicates that the proposed approach may be useful for studying and recognizing patterns associated with non-Hodgkin’s lymphomas.
Herein, we present a method that uses entropy texture signatures to distinguish chronic lymphocytic leukemia, follicular lymphoma and mantle cell lymphoma from non-Hodgkin’s lymphoma. The discriminative power of sample entropy signatures were tested via hematoxylin and eosin staining, and multiple segmentation techniques and color normalization approaches. Texture signatures were defined as feature curves obtained from multiple observations, by associating m and r parameters. The behavior of each signature was determined from areas under the curve, skewness, maximum entropy values and area ratios. The results consisted of features which were provided for different functions, such as lazy learning, trees, genetic evolution, particle swarm and animal behavior-based classifiers. The tightest associations between features and classifiers were tested as different noise levels were added. The performance achieved with the proposed method was accuracy rates between 98.72% and 99.60% with only 10 attributes, that is greater than results provided by previously established techniques, which indicates that the proposed approach may be useful for studying and recognizing patterns associated with non-Hodgkin’s lymphomas.
Descrição
Palavras-chave
Entropia amostral, Assinaturas de SampEn, Linfomas não-Hodgkin, Classificação, Segmentação, Normalização de cor, Sample entropy, SampEn signatures, Non-Hodgkin lymphomas, Classification, Segmentation, Color normalization