| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 10.42 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Introdução: A hérnia discal lombar (HDL) é, atualmente, a causa mais frequente da radiculopatia lombar nos adultos jovens. A IA é considerada “a tecnologia que define o futuro”, pelo que é extremamente pertinente demonstrar a fiabilidade do uso desta tecnologia de elevado potencial no diagnóstico de HDL, fazendo uso da técnica imagiológica de maior sensibilidade e acurácia diagnóstica, a ressonância magnética. Objetivos: O objetivo principal deste trabalho é desenvolver e treinar uma rede neuronal de convolução (CNN) destinada a auxiliar o diagnóstico de HDL, tendo por base imagens de ressonância magnética da coluna lombar no plano axial. Métodos: O desenho de estudo é de carácter descritivo e estatístico, secundário, de recuperação e análise crítica da literatura. No total, foram recolhidas e analisadas 48 345 imagens totais de ressonância magnética da coluna lombar, referentes a 515 utentes, as quais se encontram disponíveis numa base de dados pública. Destas imagens, escolheram-se 3 172 ponderadas em T2 e referentes aos planos axial e sagital. Posteriormente, recorrendo a um algoritmo de data augmentation, foram geradas 35 600 imagens desenvolver destinadas a treinar e validar duas CNN (VGG16 e VGG19). Resultados: Foram alcançados excelentes valores de accuracy durante a validação das redes, com os melhores resultados a chegarem a cerca de 0,9; estes resultados foram acompanhados de funções de loss decrescentes no processo de validação que atingiram valores de 0,5. Conclusões: O contributo deste trabalho pode ser importante para o desenvolvimento de um algoritmo capaz de detetar HDL em imagens de ressonância magnética com uma precisão muito próxima da executada pelos profissionais de saúde mais experientes.
ABSTRACT - Introduction: Lumbar disc herniation is currently the most frequent cause of lumbar radiculopathy in young adults. Artificial intelligence is considered “the technology that defines the future”, so it is extremely pertinent to demonstrate the reliability of the use of this high-potential tool in the diagnosis of lumbar disc herniations, through the imaging method of greater sensitivity and diagnostic accuracy, magnetic resonance imaging. Objectives: The main objective of this work is to demonstrate the applicability and reliability of convolutional neural networks in the diagnosis of lumbar disc hernias through the application of a convolutional neural network in magnetic resonance imaging. Methods: The study design is descriptive and statistical, secondary, recovery, and critical analysis of the literature. In total, 48 345 magnetic resonance images of the lumbar spine available in a public database were collected and analyzed, referring to 515 users. Of these images, 3 172 T2 – T2-weighted were chosen and referred to the axial and sagittal planes. Subsequently, using a data augmentation algorithm, 35,600 were selected to develop, train, and validate the CNN based on the VGG16 network. Results: Excellent accuracy values were achieved during network validation, reaching 0,9. The best loss function values in the validation process were 0,5. Conclusions: After the application of a convolutional neural network, it was found that this is a tool to be taken into account in the diagnosis of HDLs.
ABSTRACT - Introduction: Lumbar disc herniation is currently the most frequent cause of lumbar radiculopathy in young adults. Artificial intelligence is considered “the technology that defines the future”, so it is extremely pertinent to demonstrate the reliability of the use of this high-potential tool in the diagnosis of lumbar disc herniations, through the imaging method of greater sensitivity and diagnostic accuracy, magnetic resonance imaging. Objectives: The main objective of this work is to demonstrate the applicability and reliability of convolutional neural networks in the diagnosis of lumbar disc hernias through the application of a convolutional neural network in magnetic resonance imaging. Methods: The study design is descriptive and statistical, secondary, recovery, and critical analysis of the literature. In total, 48 345 magnetic resonance images of the lumbar spine available in a public database were collected and analyzed, referring to 515 users. Of these images, 3 172 T2 – T2-weighted were chosen and referred to the axial and sagittal planes. Subsequently, using a data augmentation algorithm, 35,600 were selected to develop, train, and validate the CNN based on the VGG16 network. Results: Excellent accuracy values were achieved during network validation, reaching 0,9. The best loss function values in the validation process were 0,5. Conclusions: After the application of a convolutional neural network, it was found that this is a tool to be taken into account in the diagnosis of HDLs.
Description
Mestrado em Radiações Aplicadas às Tecnologias da Saúde - Área de especialização: Imagem por Ressonância Magnética
Keywords
Ressonância magnética Hérnia discal lombar Rede neuronal de convolução Magnetic resonance imaging Lumbar Disc Herniation Convolution Neural Network
Citation
Soares SC. Desenvolvimento de uma rede neuronal de convolução para reconhecimento de hérnias discais em imagens de ressonância magnética [dissertation]. Lisboa: Escola Superior de Tecnologia da Saúde de Lisboa/Instituto Politécnico de Lisboa; 2022.
Publisher
Instituto Politécnico de Lisboa, Escola Superior de Tecnologia da Saúde de Lisboa