A data quality metric in resting fMRI and functional connectivity in mTBI

This effort furthers resting state fMRI analysis in two different areas. The first area concerns improving a common data quality metric to be more consistent across hardware parameters. The second area deals with applying techniques to resting fMRI collected from collegiate athletes in search for an objective biomarker of concussion and recovery.;While data quality is always important for fMRI, systematic changes due to factors like head motion are particularly problematic in resting fMRI where they can directly affect measures of functional connectivity. DVARS is a commonly used data quality metric based on frame to frame signal changes. Differences in hardware characteristics can cause differences in observed DVARS values, which could adversely affect the number of time points censored using a given threshold. A normalized version of DVARS called nDVARS is presented and shown to help mitigate this concern.;The second part of this work examines fMRI data collected from concussed collegiate athletes. Data were collected at baseline and at three times after injury. After appropriate preprocessing, the data were converted into a single whole-brain connectivity matrix for each time point. These matrices were used in an attempt to predict a set of clinical measures in two different ways. Forward stepwise linear regression was used with the loading scores of the top principal components as possible predictor variables, but this approach failed. Correlations were observed for each component with each of the clinical measures, revealing a significant correlation between a single component and emotional contagion after Bonferroni correction.;Each connectivity matrix was also converted into a score representing its modularity. A t-test did not find a change in modularity due to concussion. Further tests found moderate correlations between modularity and two different measures, but these did not survive Bonferroni correction for multiple comparisons. A more liberal false discovery rate correction suggests that at least one of the moderate correlations may be a true positive, but this would need to be independently validated.