Orientadores: Ivana A. Gil, Fausto Berzin

Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Odontologia de Piracicaba

Resumo: A determinação de um padrão para a atividade elétrica dos músculos mastigatórios é controversa, pois a variabilidade dos métodos de registro e as características da amostra tornam difícil a obtenção de dados eletromiográficos precisos. O uso de uma técnica de Rede Neural Artificial no...

Resumo: A determinação de um padrão para a atividade elétrica dos músculos mastigatórios é controversa, pois a variabilidade dos métodos de registro e as características da amostra tornam difícil a obtenção de dados eletromiográficos precisos. O uso de uma técnica de Rede Neural Artificial no processamento dos sinais eletromiográficos pode ser um importante instrumento para otimizar essa metodologia. O objetivo desse estudo foi investigar a possível existência de um padrão na atividade dos músculos temporais e masséteres de voluntários considerados clinicamente normais. Foram selecionadas 12 voluntárias, com idades entre 17 e 21 anos, que apresentavam ausência de sinais e sintomas de Desordens Craniomandibulares. Os sinais eletromiográficos foram captados através do eletromiógrafo Viking 11. As voluntárias foram instruídas a assumirem 3 situações mandibulares diferentes: 1. Posição de REPOUSO MANDIBULAR (R), 2. Mordida ISOTÔNICA BILATERAL (IT), e Mordida ISOMÉTRICA BILATERAL (1M). O eletromiógrafo foi calibrado numa amplitude de 200 microvolts e num tempo de 200 milisegundos. Os sinais eletromiográficos foram analisados através de um programa de Redes Neurais Artificiais (RNA), tipo Multi-Layer Perceptron, em 2 etapas: a etapa de treinamento e a etapa de testes. Os resultados da etapa de treinamento da RNA mostraram que foram atingidos os valores previstos para as três situações mandibulares
estudadas. Os resultados da etapa de teste revelaram a capacidade da RNA em reconhecer os três diferentes tipos de situações mandibulares com algum grau de acuracidade. Concluiu-se que as Redes Neurais
Artificiais podem ser utilizadas como importante ferramenta no estudo da atividade elétrica muscular, todavia a implementação das Redes Neurais Artificiais no estudo dos sinais biomédicos ainda necessita de maior pesquisa

Abstract: An electrical activity pattern for the masticatory muscle is controvertible, because the variability of record methods, different electromyographic equipment, electrical and electromagnetic interference, selection of electrodes and volunteers, to try for obtainment severa I results, that...

Abstract: An electrical activity pattern for the masticatory muscle is controvertible, because the variability of record methods, different electromyographic equipment, electrical and electromagnetic interference, selection of electrodes and volunteers, to try for obtainment severa I results, that beco me difficult establishment of real EMG data, able to represent normal electrical activity. The possibility of use an Artificial NeLiral Network (ANN) in digital processing correspond an important instrument to optimize this methodology. The aim of this study was to investigate the possible existence of a pattern in muscular activity of I Temporalis and Masseter muscle in clinically normal volunteers by using the digital processing of electromyographic signals ( Artificial Neural Network ). We selected randomly 12 female voluntears, aging 17 -21 years, with no signals and symptoms of craniomandibular disorders. The electromyographic signals was obtained by surface Beeckman electrodes, using Nicolet Electromyograph Viking 11. Ali volunteers were instructed to the obtainment three types of mandibular situation: Rest Mandibular Position ( R ), Bilateral Isotonic Bite ( IT ), Bilateral Isometric Bite ( 1M ). 200 miliseconds for time. The electromyographic signals was stored in flexible disc 3311." in ASC 11 language, transformated in DOS language by SISDIN program and that temporal arrangement allowed the analysis in Artificial Neural Network ( ANN ) program, type Multi-Layer Perceptron MLP ( Copyright @ Rational Systems, Inc, 1990-1991, Version 1.4 ), with three layers, in supervised learning; using back-propagation
algorithm, with dual exit. The analysis of electromiographic signals in ANN was divided into 2 stages: training stage and test stage. The training of ANN was realized with archives of 3 and 6 volunteers for each one of 4 muscles involved, and in the test stage we used the volunteers was not submitted to the training stage. The results of training stage of ANN showed that was reached the anticipated value for the 3 mandibular situation studied for 3 and 6 volunteers. The result of the test stage I showed the capacity of ANN by recognize the 3 different types of mandibular situation, with some degree of accuracy, and the Rest Mandibular Position was the most distinguished of the others mandibular situations. Apparently due muscle and anatomical variable, an increased sample would permit to ANN a bigger capacity of generalization ( learning ), improving the recognition of muscles activities in the bilateral isotonic and isometric bites situations. We concluded that ANN will can be used how an important tool in study of electrical activity of muscle, as well as in differential diagnosis of muscles pathologies. However, the implementation of the ANN in study of biomedical signals, require much more investigation