Welton, Thomas
(2017)
Imaging markers of brain network disruption in multiple sclerosis.
PhD thesis, University of Nottingham.
Abstract
Cognitive impairment and fatigue are prevalent and impactful symptoms of multiple sclerosis (MS). Effective markers are required by clinical studies to accurately test the efficacy of treatments for these symptoms. Graph analysis of brain networks based on magnetic resonance imaging (MRI) data can feasibly provide useful candidate markers of cognitive impairment and fatigability in MS which may be more objective, reliable and specific than existing markers. My original contribution to knowledge is therefore an exposition of the following hypothesis: “summary graph-theoretic descriptors of brain network organisation are good candidate markers of cognition or fatigue in MS”. To achieve this, network metrics were assessed based on three main criteria: reliability (“are the measurements the same across time and settings?”), validity (“do they measure what they are supposed to measure?”) and responsiveness (“are they altered when a change in cognitive state is induced?”).
The applicability of the graph-theoretic approach was first established with a spatial meta-analysis of tract integrity and its relevance to cognition and disability. Reliability over time in healthy subjects was assessed by systematic review and reliability between different scanners and between MS and control groups was assessed in two longitudinal datasets by measuring intra-class correlation (ICC) of graph metrics. The validity criterion was assessed in an analysis of covariance and linear regression of cognitive and fatigue measures with brain network metrics in people with MS. Finally, an exploration of network dynamics during a sustained attention task with a sliding-window approach was performed to test the immediate responsiveness of the measures to alterations in cognitive state.
Spatial meta-analysis of white matter tract degradation was performed using the Signed Differential Mapping method. Statistical maps were gathered from the original authors of studies which performed voxelwise correlations between fractional anisotropy (a measure of white matter integrity based on diffusion tensor imaging data) and measures of either cognitive impairment or physical disability. The combined sample included 495 people with MS and 253 controls from 12 studies. MS diagnosis was significantly associated with widespread lower tract fractional anisotropy. Distributions of voxels with significantly lower fractional anisotropy in relation to cognition and physical disability were only minimally overlapping. The number of and effect sizes for significant clusters in the cognition comparison were greater than those for the physical disability comparison, suggesting a greater relevance of cerebral white matter damage to cognition. The main results remained significant when using a stringent p-value threshold of 0.00001 to control for false positives.
The next analysis was a systematic review of the reproducibility of graph metrics over time in healthy people. Online databases were searched for articles reporting ICCs for graph metrics based on imaging data and information was recorded on the sample size, acquisition method, inter-scan interval and reported ICCs. Twenty-six articles were included, with a combined sample size of 676. Overall, reproducibility over time was rated as “good”, but heterogeneity of methods precluded in-depth quantitative analysis. A qualitative synthesis of results highlighted the main methodologic factors affecting reproducibility, which included: ICC type, retest interval, fibre tracking algorithm, graph metric type, image processing strategy, region of interest size, graph threshold and acquisition method.
Reliability of brain network metrics between scanners was tested using a travelling-subjects dataset in which 5 subjects each underwent a resting-state functional MRI scan at 10 sites. Graph metrics were calculated for each scan and then tested for ICC across sites. Reproducibility was “poor” for most metrics (characteristic path length ICC=0.23, global efficiency ICC=0.18, modularity ICC=0.24) and “fair” for two (clustering coefficient ICC=0.43, small-worldness ICC=0.42). There was limited evidence that some subjects tended to produce less reliable results and that magnets with higher field strengths did not produce more reliable results. The main implication is that multi-site studies using graph analysis of brain MRI data should investigate inter-site reproducibility beforehand.
To investigate the validity of graph metrics as markers of cognitive impairment and fatigue, MRI and neurocognitive data were first gathered from 37 people with MS and 23 matched controls. The sample was characterised in detail and comprised a range of cognitive abilities. Data quality was investigated and the small-world structure of the data was confirmed by comparison to random and lattice graphs. Analysis of covariance controlling for age, sex and education showed significant group differences for all but one graph metric. Linear regression models predicted the main measures of cognitive impairment in the MS group, but not in the control group. Measures of fatigue were not well-explained by graph metrics. The direction of the relationships indicated that greater levels of cognitive impairment were related to increased network clustering and modularity, longer average path lengths, lower small-worldness, lower levels of education, old age and sleep disturbance.
Finally, responsiveness of graph metrics was investigated in an analysis of functional network dynamics during performance of a sustained attention task. A “sliding-window” approach was taken, in which network metrics were calculated for 84 100-second windows at increments along the fMRI timeseries. Reaction times in the task showed a learning effect for both groups, but were consistently slower for the MS group. Plots of graph metrics over time showed differing responses to the task and to the transition between task and rest periods between groups. The small-worldness and clustering coefficient metrics were correlated with reaction times for both MS (small-worldness: r=0.623, <0.001; clustering coefficient: r=0.554, p=<0.001) and control (small-worldness: r=0.586, <0.001; clustering coefficient: r=0.627, p=<0.001) groups, but the characteristic path length metric was not (MS: r=-0.154, p=0.313; control: r=0.343, p=0.021).
Disconnection of cortical areas by degradation of white matter is a viable explanation for cognitive symptoms in MS. There is some evidence that increased network segregation and decreased network integration may explain cognitive symptomatology. Graph theoretic summary brain network metrics do have potential for use as complimentary information to existing markers of cognitive impairment in clinical studies.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)