Publicação
Reproducibility and sensitivity of brain network backbones: a demonstration in Small Vessel Disease
| Resumo: | Whole-brain networks have been used to study the connectivity paths within the brain, constructed using information from diffusion magnetic resonance imaging (dMRI) data and white matter fiber tractography (FT). These techniques can detect alterations in the white matter integrity and changes in axonal connections, whose alterations can be due to the presence of small vessel disease (SVD). However, there is a lack of consensus in network reconstruction methods and therefore no gold-standard model of the human brain network. Moreover, dMRI data are affected by methodological issues such as scan noise, presence of false-positive and false-negative connections. Consequently, the reproducibility and the reliability of these networks is normally very low. A potential solution to deal with the low reproducibility of brain networks is to analyze only its backbone structure. This backbone is assumed to represent the building blocks of structural brain networks and thus composed by a set of strong connections and voided of spurious connections. Such backbone should be reproducible in scan-rescan scenarios and relatively consistent between healthy subjects, while still being sensitive to disease-related changes. Several types of backbones have been proposed, constructed using white matter tractography, with dMRI data. However, no study has directly compared these backbones in terms of reproducibility, consistency, or sensitivity to disease effects in a patient population. In this project, we examined: (1) whether the proposed backbones can be applied to clinical datasets by testing if they are reproducible over two time-points and consistent between groups; (2) if they are sensitive to disease effects both in a cross-sectional and longitudinal analysis. We evaluated our research questions on a longitudinal cohort of patients with cerebral SVD and age matched controls, as well as a validation dataset of healthy young adults. Our cohort contained 87 elderly memory clinic patients with SVD recruited via the UMC Utrecht, scanned twice with an inter-scan time between baseline and follow-up of 27 ± 4 months. We also included baseline scans of 45 healthy elderly, matched in age, sex and education level, to be used as controls. Data from 44 healthy young adults was used as validation data. For each subject, we reconstructed brain structural networks from the diffusion MRI data. Subsequently, we computed 4 types of network backbone, previously described in literature: the Minimal Spanning Tree (MST), the Disparity Core, the K-Core, and Hub-Core. We compared these backbones and tested their reproducibility within subjects, and their consistency across subjects and across groups. Moreover, we performed a cross-sectional analysis between controls and patients at baseline, to evaluate if these backbones can detect disease effects and a longitudinal analysis with patient data over time, to test if they can detect disease progression. Regarding our first objective, our results show that overall MST is the backbone that shows the best reproducibility between repeated scans, as well as the highest consistency among subjects, for all of the three brain templates that we used. Secondly, the MST was also sensitive to network alterations both on a cross-sectional analysis (patients vs. controls) and on a longitudinal analysis (baseline vs. follow-up). We therefore conclude that, the use of these network backbones, as an alternative of the whole-brain network analysis, can successfully be applied to clinical datasets as a novel and reliable way to detect disease effects and evaluate disease progression. |
|---|---|
| Autores principais: | Padrela, Beatriz Esteves |
| Assunto: | Demência vascular Tractografia Substância branca Conectividade cerebral Esqueleto-neuronal Teses de mestrado - 2020 |
| Ano: | 2020 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Whole-brain networks have been used to study the connectivity paths within the brain, constructed using information from diffusion magnetic resonance imaging (dMRI) data and white matter fiber tractography (FT). These techniques can detect alterations in the white matter integrity and changes in axonal connections, whose alterations can be due to the presence of small vessel disease (SVD). However, there is a lack of consensus in network reconstruction methods and therefore no gold-standard model of the human brain network. Moreover, dMRI data are affected by methodological issues such as scan noise, presence of false-positive and false-negative connections. Consequently, the reproducibility and the reliability of these networks is normally very low. A potential solution to deal with the low reproducibility of brain networks is to analyze only its backbone structure. This backbone is assumed to represent the building blocks of structural brain networks and thus composed by a set of strong connections and voided of spurious connections. Such backbone should be reproducible in scan-rescan scenarios and relatively consistent between healthy subjects, while still being sensitive to disease-related changes. Several types of backbones have been proposed, constructed using white matter tractography, with dMRI data. However, no study has directly compared these backbones in terms of reproducibility, consistency, or sensitivity to disease effects in a patient population. In this project, we examined: (1) whether the proposed backbones can be applied to clinical datasets by testing if they are reproducible over two time-points and consistent between groups; (2) if they are sensitive to disease effects both in a cross-sectional and longitudinal analysis. We evaluated our research questions on a longitudinal cohort of patients with cerebral SVD and age matched controls, as well as a validation dataset of healthy young adults. Our cohort contained 87 elderly memory clinic patients with SVD recruited via the UMC Utrecht, scanned twice with an inter-scan time between baseline and follow-up of 27 ± 4 months. We also included baseline scans of 45 healthy elderly, matched in age, sex and education level, to be used as controls. Data from 44 healthy young adults was used as validation data. For each subject, we reconstructed brain structural networks from the diffusion MRI data. Subsequently, we computed 4 types of network backbone, previously described in literature: the Minimal Spanning Tree (MST), the Disparity Core, the K-Core, and Hub-Core. We compared these backbones and tested their reproducibility within subjects, and their consistency across subjects and across groups. Moreover, we performed a cross-sectional analysis between controls and patients at baseline, to evaluate if these backbones can detect disease effects and a longitudinal analysis with patient data over time, to test if they can detect disease progression. Regarding our first objective, our results show that overall MST is the backbone that shows the best reproducibility between repeated scans, as well as the highest consistency among subjects, for all of the three brain templates that we used. Secondly, the MST was also sensitive to network alterations both on a cross-sectional analysis (patients vs. controls) and on a longitudinal analysis (baseline vs. follow-up). We therefore conclude that, the use of these network backbones, as an alternative of the whole-brain network analysis, can successfully be applied to clinical datasets as a novel and reliable way to detect disease effects and evaluate disease progression. |
|---|