Publicação
A ecografia versus elastografia no estudo da sarcopenia
| Resumo: | Due to its effects on the health and well-being of the aging population, sarcopenia, the age-related loss in skeletal muscle mass and function, has garnered more attention. In this research, we investigate the use of cutting-edge imaging methods and machine learning approaches for the classification of sarcopenia, using ultrasound and elastography images of the rectus femoris muscle.Our study compares the performance of conventional machine learning models and cutting-edge deep learning models, evaluating each model's ability to identify sarcopenia in various age groups within a diverse dataset of 180 images collected from 30 people ranging in age from 20 to 75. The dataset underwent careful preprocessing, including data augmentation methods, to increase its diversity and potential for generalization. Neural networks, Nu Support Vector Machine, Logistic Regression, Stochastic Gradient Descent Classifier, and Support Vector Machine were all included in our investigation. Furthermore, we explored the possibilities of deep learning models such as DenseNet 121, VGG 16, VGG 19, ResNet 50, and Inception V3. Each model was adjusted and thoroughly compared based on its classification performance. Our findings show that the machine learning classifiers matched and some even surpassed the deep learning equivalents in terms of exceptional classification accuracy. Notably, the Neural Network emerged as the most reliable performer among the deep learning models, achieving a remarkable F1 score of 99.81\% in the ultrasound dataset.This outcome highlights the capability of deep learning architectures to correctly identify sarcopenia in images from elastography and ultrasound. Additionally, the neural network's competitive performance raises the possibility that high accuracy may not always require complex deep learning models, particularly when working with limited datasets.This dissertation offers insights into the classification of sarcopenia using a wide range of models, contributing to the growing body of research at the intersection of medical imaging and machine learning. The observed effectiveness of both conventional and deep learning models emphasizes the potential use of these methods in aiding medical practitioners in the early detection and intervention of sarcopenia. This finding has possible implications for enhancing the quality of life for older individuals through improved diagnostic tools and individualized care as the world's population continues to age. |
|---|---|
| Autores principais: | Lopes, Luís André Mendes |
| Assunto: | Sarcopenia Ultrasound Elastography Machine Learning Deep Learning Sarcopenia Ecografia Elastografia Machine Learning Deep Learning |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Coimbra |
| Idioma: | português |
| Origem: | Estudo Geral - Universidade de Coimbra |
| Resumo: | Due to its effects on the health and well-being of the aging population, sarcopenia, the age-related loss in skeletal muscle mass and function, has garnered more attention. In this research, we investigate the use of cutting-edge imaging methods and machine learning approaches for the classification of sarcopenia, using ultrasound and elastography images of the rectus femoris muscle.Our study compares the performance of conventional machine learning models and cutting-edge deep learning models, evaluating each model's ability to identify sarcopenia in various age groups within a diverse dataset of 180 images collected from 30 people ranging in age from 20 to 75. The dataset underwent careful preprocessing, including data augmentation methods, to increase its diversity and potential for generalization. Neural networks, Nu Support Vector Machine, Logistic Regression, Stochastic Gradient Descent Classifier, and Support Vector Machine were all included in our investigation. Furthermore, we explored the possibilities of deep learning models such as DenseNet 121, VGG 16, VGG 19, ResNet 50, and Inception V3. Each model was adjusted and thoroughly compared based on its classification performance. Our findings show that the machine learning classifiers matched and some even surpassed the deep learning equivalents in terms of exceptional classification accuracy. Notably, the Neural Network emerged as the most reliable performer among the deep learning models, achieving a remarkable F1 score of 99.81\% in the ultrasound dataset.This outcome highlights the capability of deep learning architectures to correctly identify sarcopenia in images from elastography and ultrasound. Additionally, the neural network's competitive performance raises the possibility that high accuracy may not always require complex deep learning models, particularly when working with limited datasets.This dissertation offers insights into the classification of sarcopenia using a wide range of models, contributing to the growing body of research at the intersection of medical imaging and machine learning. The observed effectiveness of both conventional and deep learning models emphasizes the potential use of these methods in aiding medical practitioners in the early detection and intervention of sarcopenia. This finding has possible implications for enhancing the quality of life for older individuals through improved diagnostic tools and individualized care as the world's population continues to age. |
|---|