Abstract :
[en] Background
{So far, brain connectivity has been successfully estimated by means of functional connectivity (FC) from functional magnetic resonance imaging (MRI), structural connectivity (SC) from diffusion weighted imaging, intersubject covariance of regional gray matter volume (GMVcov) from structural MRI, and intersubject covariance of regional glucose metabolism (FDGcov) from 18F-fluorodeoxyglucose (FDG) positron emission tomography data. To understand, in how far these estimates can be used to track physiological, i.e. task-related, and pathological, i.e., disease-specific, changes in resting state brain connectivity, data on reproducibility of these estimates are essential. Here, we determined reproducibility of group-level SC, FC, GMVcov, and FDGcov as estimated in the same subjects at rest using a simultaneous PET/MR acquisition protocol.}
Methods
Image data were acquired in 55 healthy, middle-aged individuals on a hybrid 3T PET/MR scanner. The above estimates of brain connectivity were computed from two identical acquisitions, taking place eight weeks apart, i.e., test and retest. Gray matter of each subject was parceled in native space. SC was computed as the number of streamlines connecting two GM regions and normalized by their surface area. FC was computed as Pearson correlation between time series of each pair of regions after regressing out GM and cerebrospinal fluid signals. SC and FC estimates were then averaged to obtain group-level connectivity estimates. GMVcov was computed as Pearson correlation between subject series of each pair of regions after regressing out a total GMV of each subject. Similarly, FDGcov was computed as Pearson correlation between subject series of each pair of regions after normalization to the total uptake of the GM of each subject. Reproducibility was determined using complementary measures such as (1)
Spearman’s correlation coefficient (SCC), (2) coefficient of variation (CV), and (3) proportion of connections repeatedly present in test and retest sessions.}
Results
{When considering all connections without any thresholding, high SCCs were found for all
estimates: 0.99 for SC, 0.96 for FC, 0.95 for FDGcov, and 0.93 for GMVcov. When thresholding based on sparsity and keeping only significant (p<0.05) connections, i.e., 92%<sparsity<100%, the CVs were as following: 2.7% for SC, 3.1% for FDCcov, 3.6% for GMVcov, and 5.1% for FC. When thresholding based on connectivity strength, i.e., the magnitude of the Pearson correlation coefficient as available for FC, GMVcov, and FDGcov, and keeping only the strongest connections, i.e., |0.5|<R<|1|, the relative proportion of reproducible connections was 77% for GMVcov, 56% for FDGcov, and 54% for FC. Yet, FDGcov presented the highest absolute proportion of reproducible connections (11.5%), followed by FC (2.6%), and GMVcov (0.9%).}
Conclusions
{Reproducibility is comparable among group-level SC, FC, GMcov, and FDGcov, with SC
tending to be most reproducible. Among the remaining estimates, which are derived from a statistical relationship, a substantially higher (absolute) proportion of reproducible, strong connections is found for FDGcov. Thus, FDGcov enables to explore brain connectivity in a reproducible manner over a substantially larger part of the brain than FC and GMVcov.}
