| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.15 MB | Adobe PDF |
Advisor(s)
Abstract(s)
RESUMOA voz é uma ferramenta de comunicação primordial nas relações inter-humanas, por meio de inflexões, pausas, variações de ritmo e de intensidade. É considerada a integridade da nossa identidade, pois através dela somos reconhecidos e a sua qualidade permite-nos expressar eficazmente. As patologias vocais encontram-se presentes na nossa sociedade, com profundo impacto na qualidade de vida das pessoas. A origem deve-se a várias causas e apresentam diferentes graus de gravidade. A patologia pode progredir de forma benigna ou maligna, por isso é de extrema importância ter atenção aos sinais de alteração. Um diagnóstico precoce é muito relevante para o tratamento. Porém, as formas de avaliação existentes nesta área são invasivas e desagradáveis, sendo incomodativas para o paciente. Estes aspetos motivaram o desenvolvimento de métodos não invasivos, que possam fazer uma avaliação exata e possam ser utilizados como um método de ajuda ao diagnóstico eficaz.
Neste trabalho desenvolveu-se um sistema de suporte à decisão médica no diagnóstico de patologias vocais. Para o desenvolvimento deste sistema foi necessário o estudo de um conjunto de parâmetros acústicos, bem como de classificadores, como rede neuronal artificial (RNA), com o objetivo de fazer a classificação final do paciente entre saudável e patológico. Os parâmetros utilizados neste trabalho são: Jitter Absoluto (Jitta), Jitter Relativo (Jitter), Shimmer Absoluto (ShdB), Shimmer Relativo (Shim) , Harmonic to Noise Ratio (HNR), e a Autocorrelacão. E como classificador o modelo da rede neuronal Multi Layer Perceptron (MLP).
O sistema interface gráfica desenvolvido neste trabalho servirá como um método complementar no pré-diagnóstico de patologias da voz. O modelo MLP utilizada obteve uma taxa de exatidão de 98.86% que se encontra entre os melhores valores tendo em conta estado a arte, no entanto a possibilidade da inserção deste sistema em clínicas e hospitais contribuirá para o seu aperfeiçoamento por meio de familiarização com profissionais de saúde.
Voice is a powerful tool that make possible the communication between humans being, by way of inflections, pauses, variations in rhythm and intensity. As the integrity of our identity, it is considered to allow us to express ourselves effectively to be recognized. Vocal pathology become a reality in modern society, having a deep impact by affecting people life quality. it is extremely important to be aware of signs of alterations that can be derived from different causes, with different degrees of severity, which evolve in a benign or malignant. Early diagnostic intervention revealed to be extremely crucial to treatment. However, the already existed examination procedures in this field show to be invasive and unpleasant, which can be uncomfortable to the patient. Therefore, the need to develop a non-invasive method emerged in order to delivery an accurate and effective diagnosis. In this work, a medical decision support system in the diagnosis of vocal pathologies was developed. For the development of this system, it was necessary to study a set of acoustic parameters, as well as classifiers, of artificial neural networks (ANN), with the objective of making the final classification of the patient between healthy and pathological. The parameters used in this work are Absolute Jitter (Jitta), Relative Jitter (Jitt), Absolute Shimmer (ShdB), Relative Shimmer (Shim), Harmonic to Noise Ratio (HNR), and Autocorrelation. And as a classifier, the Multi-Layer Perception (MLP) neural network model. The graphic interface system developed in this work will serve as a complementary method in the pre-diagnosis of voice pathologies. The MLP model used obtained a success rate of 98.86%, which is among the best values taking into account the state of the art, however the possibility of inserting this system in clinics and hospitals will contribute to its improvement through familiarization with health professionals.
Voice is a powerful tool that make possible the communication between humans being, by way of inflections, pauses, variations in rhythm and intensity. As the integrity of our identity, it is considered to allow us to express ourselves effectively to be recognized. Vocal pathology become a reality in modern society, having a deep impact by affecting people life quality. it is extremely important to be aware of signs of alterations that can be derived from different causes, with different degrees of severity, which evolve in a benign or malignant. Early diagnostic intervention revealed to be extremely crucial to treatment. However, the already existed examination procedures in this field show to be invasive and unpleasant, which can be uncomfortable to the patient. Therefore, the need to develop a non-invasive method emerged in order to delivery an accurate and effective diagnosis. In this work, a medical decision support system in the diagnosis of vocal pathologies was developed. For the development of this system, it was necessary to study a set of acoustic parameters, as well as classifiers, of artificial neural networks (ANN), with the objective of making the final classification of the patient between healthy and pathological. The parameters used in this work are Absolute Jitter (Jitta), Relative Jitter (Jitt), Absolute Shimmer (ShdB), Relative Shimmer (Shim), Harmonic to Noise Ratio (HNR), and Autocorrelation. And as a classifier, the Multi-Layer Perception (MLP) neural network model. The graphic interface system developed in this work will serve as a complementary method in the pre-diagnosis of voice pathologies. The MLP model used obtained a success rate of 98.86%, which is among the best values taking into account the state of the art, however the possibility of inserting this system in clinics and hospitals will contribute to its improvement through familiarization with health professionals.
Description
Mestrado em IPB-ESTG
Keywords
Patologias da voz Sinal da fala Sistema Parâmetros acústicos Redes neuronais