Publicação
Automatic classification of medical images based on functional connectivity measurements - methodological exploration
| Resumo: | The study of patterns of neuronal activity constitutes a tool of extreme value in the attempt to unveil neural pathological mechanisms. Hence, functional connectivity studies using images from Resting State fMRI (rs-fMRI) are crucial, and there are several metrics which can be used to assess brain connections. Nonetheless, no clear evidence exists that some may be better than others. In this study, in an attempt to discover if certain metrics better characterized certain connections, two different approaches were followed. Data from a public dataset was used - Addiction Connectome Preprocessed Initiative (ACPI) - as well as one toolbox for matrix construction - Multiple Connectivity Analysis (MULAN) - and another for statistical comparison - GraphVar. Both toolboxes run in MATLAB. Metrics under analysis were: correlation, coherence, mutual information, transfer entropy and non-linear correlation. To that end, 116 brain regions were considered. First, considering only healthy subjects, it was done a pairwise comparison between results from different metrics. It was verified that each of them led to different results regarding the same connections. Then, connectivity results between a healthy and a pathological group of subjects with Attention-Deficit/Hyperactivity Disorder (ADHD) were compared. Concerning the differences, several similarities with the known affected areas described amongst the literature were found. However, discrepancies were observed which may be related to differences in sample size and/or the metric used in these studies. In general, it was shown that there is indeed variability between functional metrics and regional specificity. Still, the anatomical and physiological reasons for these differences remain unknown. It was clear that using more than one metric may be important and that the use of more general metrics may have advantages in the study of the pathological brain as it may have more complex dynamics. Furthermore, ensemble tools that have into consideration more than one metric to characterize brain connections may represent invaluable tools for autonomic image classification. |
|---|---|
| Autores principais: | Ramos, Vanessa Gouveia |
| Assunto: | Functional connectivity rs-fMRI ADHD functional connectivity metrics regional specificity |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | The study of patterns of neuronal activity constitutes a tool of extreme value in the attempt to unveil neural pathological mechanisms. Hence, functional connectivity studies using images from Resting State fMRI (rs-fMRI) are crucial, and there are several metrics which can be used to assess brain connections. Nonetheless, no clear evidence exists that some may be better than others. In this study, in an attempt to discover if certain metrics better characterized certain connections, two different approaches were followed. Data from a public dataset was used - Addiction Connectome Preprocessed Initiative (ACPI) - as well as one toolbox for matrix construction - Multiple Connectivity Analysis (MULAN) - and another for statistical comparison - GraphVar. Both toolboxes run in MATLAB. Metrics under analysis were: correlation, coherence, mutual information, transfer entropy and non-linear correlation. To that end, 116 brain regions were considered. First, considering only healthy subjects, it was done a pairwise comparison between results from different metrics. It was verified that each of them led to different results regarding the same connections. Then, connectivity results between a healthy and a pathological group of subjects with Attention-Deficit/Hyperactivity Disorder (ADHD) were compared. Concerning the differences, several similarities with the known affected areas described amongst the literature were found. However, discrepancies were observed which may be related to differences in sample size and/or the metric used in these studies. In general, it was shown that there is indeed variability between functional metrics and regional specificity. Still, the anatomical and physiological reasons for these differences remain unknown. It was clear that using more than one metric may be important and that the use of more general metrics may have advantages in the study of the pathological brain as it may have more complex dynamics. Furthermore, ensemble tools that have into consideration more than one metric to characterize brain connections may represent invaluable tools for autonomic image classification. |
|---|