• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.18.2006.tde-27112006-083144
Document
Author
Full name
Patricia Bellin Ribeiro
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2006
Supervisor
Committee
Schiabel, Homero (President)
Gutierrez, Marco Antonio
Romero, Roseli Aparecida Francelin
Title in Portuguese
Classificação por análise de contornos de nódulos mamários utilizando redes neurais artificiais
Keywords in Portuguese
câncer de mama
classificação
mamografia
redes neurais
textura
Abstract in Portuguese
Este trabalho apresenta a proposta de uma metodologia para classificação de nódulos mamários por contorno. O contorno do nódulo apresenta difícil interpretação pelos especialistas, devido à dificuldade de vizibilização e o baixo contraste das imagens mamográficas. As imagens utilizadas foram obtidas do banco de imagens mamograficas do LAPIMO, no total foram utilizadas 135 imagens contendo laudos por contorno. Através das imagens mamográficas digitalizadas são recortadas região de interesse (RI) de onde serão extraídos descritores de textura, intensidade e geométricos com o objetivo de caracterizar os padrões de contorno de nódulos. Após a extração desses descritores foram utilizados dois métodos de seleção de atributos, um utilizando rede neural self-organizing map (SOM) e Gaussianas e outro utilizando matriz de covariância. Os atributos extraídos serviram de entrada para duas redes neurais a multi-layer perceptron (MLP) e SOM, através do qual, foram realizados diversos treinamentos utilizando diferentes entradas, várias topologias e diferentes saídas, devido às várias combinações de classes. Dentre todos os treinamentos realizados, o treinamento escolhido para compor o classificador final foi o conjunto formado pelas 5 classes, obtido pela rede neural MLP com topologia de 20 neurônios de entrada, 40 neurônios na primeira camada intermediária, 20 neurônios na segunda cama intermediária e 5 neurônios na camada de saída, com taxa de aprendizagem igual a 0,9 e erro menor que 0,01, as 20 entradas foram selecionadas através da rede SOM e Gaussianas. O acerto obtido utilizando 135 RIs e pesos fixos foi de 89% de acerto total, Az igual a 0,98, falso negativo igual a 5% e falso positivo igual a 7%. O classificador apresentado nesse trabalho será acrescentado ao classificador já existente no esquema CAD em mamografia.
Title in English
Classification by analysis of contour of mammary masses using artificial neural networks
Keywords in English
breast cancer
classification
mammography
neural networks
texture
Abstract in English
This work presents the proposal of a methodology for mammary nodules classification by their contour. The nodule bounder is had to interpret due to the low contrast of the mammographic images. Images used in the test obtained from the LAPIMO’s database a 135 of images, with information about the masses. From regions of interest (RÓIS) selected in such images texture, intensity and geometric features were extracted in order to characterize the patterns of nodules contour. After the features extraction two methods of features selection, were used: one using self-organizing map (SOM) neural networks together Gaussian distributions analysis and another using covariance matrix. The extracted features were the input for two neural networks, a multi-layer perceptron (MLP) and a SOM. Several trainings were accomplished using different inputs, several topologies and different exits, regarding several classes combinations. Among all of the trainings, the choice has corresponded to the group formed by 5 classes, obtained for the MLP neural networks in a topology of 20 input neurons, 40 neurons in the first intermediate layer, 20 neurons in the second intermediate layer and 5 neurons in the exit layer, with learning rate of 0.9 and error smaller than 0.01. The 20 inputs were selected by the SOM network together Gaussian distributions. The performance using 135 ROIs and fixed weights has registred 89% of right responses with Az = 0.98, false negative rate of 5% and false positive rate of 7%. The classifier presented in this work is being implemented in the general classifier as part of our group CAD scheme in mammography.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Patricia.pdf (3.04 Mbytes)
Publishing Date
2006-12-01
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.