• 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.17.2023.tde-05012024-162435
Document
Author
Full name
Luan Oliveira Barreto
Institute/School/College
Knowledge Area
Date of Defense
Published
Ribeirão Preto, 2023
Supervisor
Committee
Muglia, Valdair Francisco (President)
Colli, Leandro Machado
Macedo, Tulio Augusto Alves
Title in Portuguese
Capacidade de predição de agressividade de neoplasias renais malignas utilizando radiômica em sequências específicas de ressonância magnética
Keywords in Portuguese
Inteligência artificial
Logistic regression
Neoplasia renal
PyRadiomics
Random forest
Ressonância magnética
Sklearn
SVM
Abstract in Portuguese
Objetivo: Avaliar a capacidade de sequências específicas de RM em predizer a gravidade anatomopatológica de neoplasias renais malignas, utilizando o desempenho da radiômica, baseada em ferramentas de processamento por inteligência artificial com aprendizado de máquina (machine learning), treinada com base em exames de pacientes com confirmação histopatológica de câncer de rim. Materiais e métodos: Segmentamos, manualmente, 42 lesões renais de exames de RM, realizadas em nossa instituição. Os dados foram adquiridos nas fases T2, ADC e Portal, através do software 3D Slicer. Em seguida, foram extraídos atributos (features) das imagens usando a biblioteca PyRadiomics. Após a extração dos atributos, realizou-se uma seleção de atributos utilizando a biblioteca Sklearn. Foram testados diferentes modelos de seleção, como Recursive Feature Elimination (RFE) e SelectKBest. Os modelos de aprendizado de máquina testados foram Random Forest, Sector Vector Machine (SVM) e Logistic Regression. O treino foi feito com a porção de treino dos casos, e a validação dos modelos com a porção de teste. Após a validação dos modelos, estudamos a acurácia, a precisão e a área sob a curva ROC (AUC-ROC) para avaliar o desempenho dos modelos. Resultados: no ADC, o SVM, Random Forest e Logistic Regression alcançaram acurácia de 0.80, 0.68 e 0.78, precisão de 0.65, 0.46 e 0.59 e AUC-ROC de 0.73, 0.57 e 0.73. No Portal, o SVM, Random Forest e Logistic Regression alcançaram acurácia de 0.82, 0.83 e 0.78, precisão de 0.69, 0.62 e 0.64 e AUC-ROC de 1.00, 0.67 e 0.82. Discussão: os resultados deste estudo demonstram que a radiômica pode ser usada para diferenciar entre tumores renais de baixo e alto grau usando dados de ressonância magnética com boa precisão. O SVM e Logistic Regression obtiveram bons resultados, com SVM alcançando a maior precisão e o Random Forest não obteve bons resultados. Novos estudos com amostras maiores são necessários para confirmar esses achados.
Title in English
Ability to predict the anatomopathological severity of malignant renal neoplasms using radiomics in specific sequences of magnetic resonance
Keywords in English
Artificial intelligence
Kidney neoplasm
Logistic regression
Magnetic resonance
PyRadiomics
Random forest
Sklearn
SVM
Abstract in English
Objective: To evaluate the ability of specific MRI sequences to predict the anatomopathological severity of malignant renal neoplasms, using the performance of radiomics, based on processing tools by artificial intelligence with machine learning, trained on the basis of patient exams with histopathological confirmation of kidney cancer. Materials and methods: We manually segmented 42 renal lesions from MRI scans performed at our institution. Data were acquired in phases T2, ADC and Portal, through the 3D Slicer software. Then, features of the images were extracted using the PyRadiomics library. After extracting the attributes, a selection of attributes was performed using the Sklearn library. Different selection models were tested, such as Recursive Feature Elimination (RFE) and SelectKBest. The tested machine learning models were Random Forest, Sector Vector Machine (SVM) and Logistic Regression. Training was done with the training portion of the cases, and validation of the models with the test portion. After validating the models, we studied the accuracy, precision and area under the ROC curve (AUC-ROC) to evaluate the performance of the models. Results: in the ADC, the SVM, Random Forest and Logistic Regression achieved accuracy of 0.80, 0.68 and 0.78, precision of 0.65, 0.46 and 0.59 and AUC-ROC of 0.73, 0.57 and 0.73. In Portal, SVM, Random Forest and Logistic Regression reached accuracy of 0.82, 0.83 and 0.78, precision of 0.69, 0.62 and 0.64 and AUC-ROC of 1.00, 0.67 and 0.82. Discussion: The results of this study demonstrate that radiomics can be used to differentiate between low and high grade renal tumors using MRI data with good accuracy. SVM and Logistic Regression achieved good results, with SVM achieving the highest accuracy and Random Forest not achieving good results. New studies with larger samples are needed to confirm these findings.
 
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.
LuanOliveiraBarreto.pdf (940.17 Kbytes)
Publishing Date
2024-03-12
 
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.