Abstract
2427
Introduction: Assessment of test-retest reproducibility is essential to differentiate changes in a measurement of interest over time from variability due to technical and biological noise. In the field of brain connectivity, magnetic resonance imaging (MRI) techniques such as functional MRI (fMRI), diffusion weighted imaging (DWI), and structural MRI (sMRI) are commonly used to provide measures of functional connectivity (FC), structural connectivity (SC) and grey matter volume covariance (GMVcov), respectively. There is also increasing evidence that positron emission tomography (PET) with 18F-fluorodeoxyglucose (FDG), is also able to provide a measure of brain FC. Unlike fMRI-FC, which is based on a statistical relationship of blood oxygen level dependent (BOLD) signal measured over time, PET-FC relies on covariance of regional FDG uptake across subjects (FDGcov). While FDGcov is increasingly used in the field of brain connectivity, its test-retest reproducibility remains unknown. Here, we quantified test-retest reproducibility of FDGcov relative to MRI-derived established measures of brain connectivity.
Methods: Multimodal imaging was performed in 55 healthy individuals (56±4 years; 24 female) on a hybrid PET/MR scanner twice, 8 weeks apart. FSL, MATLAB v2020a, SPM12 and ANTs tools were used to process resting state fMRI, DWI, sMRI, and FDG-PET data as well as to construct group-based networks of FC, SC, GMVcov, and FDGcov. First, GM of each subject was segmented into 106 regions, including cortex, cerebellum and subcortical structures. FC was measured through Pearson correlations of BOLD signals between regions. Potential white matter fibers in terms of SC were reconstructed from DWI data using probabilistic tractography. The number of streamlines between the regions served as measure of SC. Pearson correlations of regional grey matter volume across subjects were computed to build a network of GMVcov. FDGcov was computed as inter-regional Pearson correlations of normalized FDG uptake across subjects. We examined reproducibility of these networks by calculating the Spearman correlation coefficient between the test and re-test connection weights. In addition, we computed the coefficient of variance (CV), the ratio of standard deviation of test and retest values to the mean. CV is a statistical measure of the dispersion of data points in a data series around the mean; the lower the absolute value of the CV (abs(CV)), the more precise the estimate. Specifically, we calculated the median abs(CV) of the networks at the same sparsity level focusing on 8% strongest connections (i.e. sparsity >92%). Then, we quantified the averaged CV through the area under the curve (AUC) normalized by the sparsity range. Finally, we compared the histograms of cumulative frequency for abs(CV) between different modalities.
Results: The Spearman correlation coefficients were high for all modalities, being maximum for SC (0.99), followed by FC (0.96), FDGcov (0.95), and GMVcov (0.93).
Across all measures, weaker connections showed higher CV. SC showed substantially lower CV than all other measures. Specifically, the averaged median abs(CV) was 0.23% for SC; 3.4% for FDGcov, 3.8% for GMVcov and 5.2% for FC.
Finally, histograms of cumulative frequency revealed that the proportion of connections with a good reproducibility (CV<10%) is 67% for SC, 29% for FC and FDGcov, and 27% for and GMVcov.
Conclusions: Despite the differences in calculation, FDGcov has an overall test-retest reproducibility similar to that of the established fMRI-FC. These support that FDGcov as sovereign index of brain functional connectivity.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.