| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 13.84 MB | Adobe PDF |
Authors
Abstract(s)
As imagens por ressonância magnética são utilizadas no apoio ao diagnóstico de doenças neurológicas, como a doença de Alzheimer. Por se tratar de um exame com diversas sequências, muito demorado, têm-se desenvolvido diferentes métodos para que a aquisição da imagem seja mais rápida. Contudo, existem dúvidas sobre o seu impacto no diagnóstico, em particular na sua utilização com algoritmos de inteligência artificial. Dado o contexto, o objetivo principal desta investigação é comparar o desempenho de imagens RM ponderadas em T1 aceleradas ou não aceleradas na capacidade discriminativa da doença de Alzheimer. Os dados utilizados foram imagens de ressonância magnética, exclusivamente obtidos através da plataforma on-line ADNI. Foram utilizadas imagens de sequências MPRAGE ponderadas em T1 3D adquiridas em 3T. A amostra utilizada foi 73 indivíduos com doença de Alzheimer e 73 cognitivamente normais, cada uma delas com imagens não aceleradas e aceleradas. O processamento das imagens foi realizado com o programa FreeSurfer para parcelização em volumes cerebrais. Os volumes de imagens aceleradas e não aceleradas foram comparados estatisticamente fazendo recurso ao programa SPSS. Observou-se diferenças significativas dos volumes cerebrais obtidos das imagens aceleradas estando aumentados nos doentes e diminuídos nos cognitivos normais. Foram utilizados ainda algoritmos de análise de discriminante linear e análise de regressão logística. Observou-se uma boa capacidade de discriminação entre os doentes e saudáveis, sendo contudo menor aquando da utilização das imagens aceleradas. Neste caso, as diferenças encontradas nos volumes afetaram a exatidão da discriminação. Sugere-se, portanto, que em análises futuras não se devem misturar imagens adquiridas com ou sem aceleração sob o risco de se comprometerem os resultados.
ABSTRACT - Magnetic resonance imaging is used to support the diagnosis of neurological diseases such as Alzheimer's disease. Because it is an exam with several sequences, which takes a long time, different methods have been developed to make image acquisition faster. However, there are doubts about its impact on diagnosis, in particular its use with artificial intelligence algorithms. Given the context, the main objective of this investigation was to compare the performance of accelerated and non-accelerated T1-weighted MR images regarding their ability to discriminate Alzheimer's disease. The data used were magnetic resonance images, exclusively obtained through the ADNI online platform. Images of 3D T1-weighted MPRAGE sequences acquired in 3T were used. The sample used was 73 individuals with Alzheimer's disease and another 73 cognitively normal individuals, each with non-accelerated and accelerated images. Image processing was performed using the FreeSurfer program for brain volume parcelization. The volumes of accelerated and non-accelerated images were statistically compared using the SPSS program. Significant differences were observed in the cerebral volumes obtained from the accelerated images, being increased in patients and decreased in cognitively normal individuals. Linear discriminant analysis and logistic regression analysis were also used. A good ability to discriminate between patients and cognitively normal individuals was observed, however, it was lower when using accelerated images. In this case, the differences found in the volumes affected the discrimination accuracy. Therefore, it is suggested that, in future analyses, images acquired with or without acceleration should not be mixed under the risk of compromising the results.
ABSTRACT - Magnetic resonance imaging is used to support the diagnosis of neurological diseases such as Alzheimer's disease. Because it is an exam with several sequences, which takes a long time, different methods have been developed to make image acquisition faster. However, there are doubts about its impact on diagnosis, in particular its use with artificial intelligence algorithms. Given the context, the main objective of this investigation was to compare the performance of accelerated and non-accelerated T1-weighted MR images regarding their ability to discriminate Alzheimer's disease. The data used were magnetic resonance images, exclusively obtained through the ADNI online platform. Images of 3D T1-weighted MPRAGE sequences acquired in 3T were used. The sample used was 73 individuals with Alzheimer's disease and another 73 cognitively normal individuals, each with non-accelerated and accelerated images. Image processing was performed using the FreeSurfer program for brain volume parcelization. The volumes of accelerated and non-accelerated images were statistically compared using the SPSS program. Significant differences were observed in the cerebral volumes obtained from the accelerated images, being increased in patients and decreased in cognitively normal individuals. Linear discriminant analysis and logistic regression analysis were also used. A good ability to discriminate between patients and cognitively normal individuals was observed, however, it was lower when using accelerated images. In this case, the differences found in the volumes affected the discrimination accuracy. Therefore, it is suggested that, in future analyses, images acquired with or without acceleration should not be mixed under the risk of compromising the results.
Description
Mestrado em Radiações Aplicadas às Tecnologias da Saúde. - Área de especialização: Imagem por Ressonância Magnética
Keywords
Ressonância magnética Doença de Alzheimer Aquisição acelerada Aquisição em paralelo Imagem volumétrica Classificação automática Magnetic resonance Alzheimer disease Accelerated acquisition Parallel acquisition Volumetric images Machine-learning classification
Citation
Cechinel CM. Efeito da aceleração na aquisição de imagem ponderada em T1 volumétrica na classificação de doentes com Alzheimer [dissertation]. Lisboa: Escola Superior de Tecnologia da Saúde de Lisboa/Instituto Politécnico de Lisboa; 2022.