• 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.59.2023.tde-20062023-152537
Document
Author
Full name
Sergio Baldo Júnior
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
Ribeirão Preto, 2023
Supervisor
Committee
Tinós, Renato (President)
Carneiro, Murillo Guimarães
Rosa, João Luis Garcia
Title in Portuguese
Algoritmos genéticos e aprendizado profundo baseado em redes neurais recorrentes do tipo LSTM para auxílio ao diagnóstico médico
Keywords in Portuguese
Acidente vascular cerebral
Algoritmo genético
Coma
Convolutional neural networks
Eletroencefalograma
Long short term memory
Abstract in Portuguese
Exames de Eletroencefalograma (EEG) têm se tornado uma ferramenta essencial no diagnóstico e avaliação de diversas doenças neurológicas. A análise automática de exames de EEG por meio de algoritmos computacionais é uma importante ferramenta para auxiliar médicos especialistas no diagnóstico mais preciso dessas doenças. Nesse sentido, neste trabalho, é proposto um modelo híbrido de Aprendizado Profundo baseado em Redes Neurais Recorrentes do tipo LSTM (Long Short Term Memory) para auxiliar na análise de sinais de EEG. O modelo híbrido proposto utiliza informações de duas diferentes fontes: i) camadas intermediárias do modelo CNN (Convolutional Neural Network) - LSTM; ii) características adicionais relacionadas a informações dos pacientes e extraídas de sinais de EEG. As saídas da camada LSTM e as características adicionais são inseridas como entradas para a primeira camada densa do modelo CNN - LSTM. O modelo é testado em sinais de EEG de pacientes em Coma e pacientes que sofreram Acidente Vascular Cerebral. Além disso, um Algoritmo Genético é utilizado para selecionar o melhor subconjunto de características e otimizar os hiper-parâmetros do modelo híbrido proposto. Os resultados dos experimentos sugerem que a inclusão de características relacionadas a informações dos pacientes e extraídas de sinais de EEG potencializou o desempenho do classificador. No entanto, nem todas as características contribuíram para aumentar o desempenho do classificador. A utilização do Algoritmo Genético na seleção das características mais relevantes foi capaz de produzir um modelo com resultados superiores aos modelos utilizados como referência. Adicionalmente, o Algoritmo Genético é capaz de encontrar a melhor arquitetura híbrida CNN - LSTM para classificar os sinais de EEG em cada base de dados testada. O modelo proposto apresenta novas possibilidades para auxiliar médicos na análise de sinais de EEG e no prognóstico e tratamento dos pacientes.
Title in English
Genetic algorithms and deep learning based on recurrent neural networks of the LSTM type to aid medical diagnostic
Keywords in English
Coma
Convolutional neural network
Electroencephalogram
Genetic algorithm
Long short term memory
Stroke
Abstract in English
In recent years, Electroencephalogram (EEG) exams have become an essential tool in the diagnosis and evaluation of various neurological diseases. Automatic analysis of EEG exams through computational algorithms is an important tool to assist medical experts in the more precise diagnosis of these diseases. In this work, we propose a hybrid Deep Learning model based on Long Short-Term Memory (LSTM) Recurrent Neural Network to aid in the analysis of EEG signals. The proposed hybrid model uses information from two different sources: i) intermediate layers of the Convolutional Neural Network (CNN) - LSTM model; ii) additional features related to patient information and extracted from EEG signals. The outputs of the LSTM layer and additional features are inserted as inputs to the first dense layer of the CNN - LSTM model. The model is tested on EEG signals from Coma patients and patients who have suffered a Stroke. Additionally, a Genetic Algorithm is used to select the best subset of features and optimize the hyper-parameters of the proposed hybrid model. The experimental results suggest that the inclusion of features related to patient information and extracted from EEG signals enhanced the classifier's performance. However, not all features contributed to increasing the classifier's performance. The use of the Genetic Algorithm in selecting the most relevant features was able to produce a model with superior results to the reference models used. Furthermore, the Genetic Algorithm is capable of finding the best CNN - LSTM hybrid architecture to classify EEG signals in each tested database. The proposed model presents new possibilities to assist medical professionals in the analysis of EEG signals and the prognosis and treatment of patients.
 
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.
Publishing Date
2023-06-20
 
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.