Logo Logo
Help
Contact
Switch language to German
Univariate and multivariate pattern analysis of preterm subjects: a multimodal neuroimaging study
Univariate and multivariate pattern analysis of preterm subjects: a multimodal neuroimaging study
Background: Widespread lasting functional connectivity (FC) and brain volume changes in cortices and subcortices after premature birth have been researched in recent studies. However, the relationship remains unclear between spontaneously slow blood oxygen dependent level (BOLD) fluctuations and gray matter volume (GMV) changes in specific brain areas, such as temporal insular cortices, and whether classification methods based on MRI could be successfully applied to the identification of preterm individuals. In this thesis I hypothesized that in prematurely born adults 1. Ongoing neural excitability and brain activity, as estimated by regional functional connectivity of resting state functional MRI (rs-fMRI) is accompanied with altered low-frequency fluctuations and neonatal complications; 2. Altered regional functional connectivity is connected with superimposed cerebral structural reductions; and 3. multivariate neuroanatomical and functional brain patterns could be treated as features to identify preterm subjects from term subjects individually. Methods: To investigate these hypotheses, the principal results of structural alterations were measured with voxel-based morphometry (VBM), while rs-fMRI outcomes were estimated with amplitude of low-frequency fluctuations (ALFF) in analysis with ninety-four very preterm/very low birth weight (VP/VLBW) and ninety-two full-term (FT) born young adults. Results: The results of the thesis support the hypotheses by showing that, in univariate results, first in VP/VLBW grownups, ALFF was decreased in the left lateral temporal cortices no matter with global signal regression, and this reduction was closely associated with neonatal complications and cognitive variables. Second overlapped brain regions were found between reduced ALFF and reduced brain volumes in the left temporal cortices, and positively associated with each other, demonstrating a potential relationship between VBM and ALFF in this brain area. In multimodal multivariate pattern recognition analysis (MVPA), the gray matter volume (GMV) classifier displayed a higher accuracy (80.7%) contrast with the ALFF classifier (77.4%). The late fusion of GMV and ALFF did not outperform single GMV modality classification by reaching 80.4% accuracy. Moderator analysis from both rs-fMRI and structural MRI (sMRI) uncovered that the neuro-prematurity performance was predominantly determined by neonatal complications. Conclusions: In conclusion, these outcomes exhibit the long term effects of premature labour on lateral temporal cortices, which changed in both ongoing BOLD fluctuations and decreased cerebral structural volumes. This thesis further provided evidence that multivariate pattern analysis such as support vector machine (SVM) may identify imaging-based biomarkers and reliably detect signatures of preterm birth.
Not available
Shang, Jing
2019
English
Universitätsbibliothek der Ludwig-Maximilians-Universität München
Shang, Jing (2019): Univariate and multivariate pattern analysis of preterm subjects: a multimodal neuroimaging study. Dissertation, LMU München: Faculty of Medicine
[img]
Preview
PDF
Shang_Jing.pdf

2MB

Abstract

Background: Widespread lasting functional connectivity (FC) and brain volume changes in cortices and subcortices after premature birth have been researched in recent studies. However, the relationship remains unclear between spontaneously slow blood oxygen dependent level (BOLD) fluctuations and gray matter volume (GMV) changes in specific brain areas, such as temporal insular cortices, and whether classification methods based on MRI could be successfully applied to the identification of preterm individuals. In this thesis I hypothesized that in prematurely born adults 1. Ongoing neural excitability and brain activity, as estimated by regional functional connectivity of resting state functional MRI (rs-fMRI) is accompanied with altered low-frequency fluctuations and neonatal complications; 2. Altered regional functional connectivity is connected with superimposed cerebral structural reductions; and 3. multivariate neuroanatomical and functional brain patterns could be treated as features to identify preterm subjects from term subjects individually. Methods: To investigate these hypotheses, the principal results of structural alterations were measured with voxel-based morphometry (VBM), while rs-fMRI outcomes were estimated with amplitude of low-frequency fluctuations (ALFF) in analysis with ninety-four very preterm/very low birth weight (VP/VLBW) and ninety-two full-term (FT) born young adults. Results: The results of the thesis support the hypotheses by showing that, in univariate results, first in VP/VLBW grownups, ALFF was decreased in the left lateral temporal cortices no matter with global signal regression, and this reduction was closely associated with neonatal complications and cognitive variables. Second overlapped brain regions were found between reduced ALFF and reduced brain volumes in the left temporal cortices, and positively associated with each other, demonstrating a potential relationship between VBM and ALFF in this brain area. In multimodal multivariate pattern recognition analysis (MVPA), the gray matter volume (GMV) classifier displayed a higher accuracy (80.7%) contrast with the ALFF classifier (77.4%). The late fusion of GMV and ALFF did not outperform single GMV modality classification by reaching 80.4% accuracy. Moderator analysis from both rs-fMRI and structural MRI (sMRI) uncovered that the neuro-prematurity performance was predominantly determined by neonatal complications. Conclusions: In conclusion, these outcomes exhibit the long term effects of premature labour on lateral temporal cortices, which changed in both ongoing BOLD fluctuations and decreased cerebral structural volumes. This thesis further provided evidence that multivariate pattern analysis such as support vector machine (SVM) may identify imaging-based biomarkers and reliably detect signatures of preterm birth.