Table 1.
Summary of Included Studies
| Study | n | Interval | Acquisition | Conclusions |
|---|---|---|---|---|
| Andreotti | 19 | <1 hour | DTI | Smaller ROIs produced less reliable graph metrics. |
| Bassett | 6 | 5.5 days | DTI, DSI | Metrics were more reliable based on DTI data compared with DSI data: DSI is too susceptible to noise. Metrics were more reliable based on structural atlases compared with functional atlases. Larger ROIs produced more reliable graph metrics. Correcting edge weights for ROI size did not improve reliability. |
| Braun | 33 | 14 days | RS-fMRI | Reducing the length of the time series and global signal regression did not improve reliability. Using the first eigenvariate and the median time series instead of the mean did not improve reliability. Using a broader frequency range significantly improved reliability. |
| Buchanan | 10 | 2–3 days | DTI | Networks based on white matter ROIs proved more reliable than gray matter ROIs. Using a deterministic or probabilistic tractography algorithm or varying the edge weight definition did not significantly affect reliability. |
| Cao | 26 | 14±2.1 days | RS-fMRI, n-back fMRI | Global metrics were more reliable than local metrics. Coarse structural atlases were less reliable than finer parcellation schemes. |
| Cheng | 44 | 1 week | DTI | Reliability was sensitive to the two different edge weighting schemes tested. |
| Dennis | 17 | 101±18 days | DTI | Most metrics were reliable above densities of 0.3. |
| Deuker | 16 | 4–6 weeks | RS-MEG, n-back MEG | Reliability was higher during performance of a task. Task practice was associated with reliability. Metrics were more reliable in low-frequency bands. |
| Duda | 21 | <1 hour | DTI | Choice of deterministic tracking algorithm did not significantly affect intrasubject reliability. Between subjects, some metrics are more sensitive than others to the choice of deterministic tracking algorithm. Choice of the anatomical label set affects some metric scores and reliabilities. |
| Fan | 16 | 1 year | RS-fMRI | Having eyes open or closed during rest affected reliability of graph metrics. Hubs were less sensitive to changes in rest condition. |
| Guo | 24 | 13±3 months | RS-fMRI | Wavelet transformation of ROI time series improved reliability. Reliability of betweenness centrality was poor compared to degree and clustering coefficient. Soft and proportional thresholding did not improve reliability. |
| Jin | 10 | 15±8.4 days | RS-MEG | Reliability varies depending on eyes open and closed states, frequency band, metric type, and MEG sensor position. High frequencies were less reliable. |
| Liang | 22, 25 | <1 hour, 11±4 months | RS-fMRI | Demonstration of differences in reliability with different image processing strategies (Pearson vs. partial correlation, global signal regression, and frequency band). |
| Liao | 11 | 1 week | RS-fMRI | Degree centrality was more reliable at longer scan durations. Degree centrality was more reliable in some brain regions with a shorter TR. Global signal regression reduced reliabilities. |
| Niu | 21 | 20 minutes | RS-fNIRS | Different hemoglobin concentration types produced different graph metric reliabilities. Denoising the data with ICA did not improve reliability. |
| Owen | 10, 5 | 60.8±33.6 days | DTI | Metrics were more reliable using individual segmentation than when using a high-resolution atlas. Number of streamlines per voxel and weighting edges by connection strength did not affect metrics' reliability. |
| Park | 12 | 3 hours (×8) | RS-fMRI | Local efficiency is more reliable than global efficiency. |
| Parker | 28 | <1 hour | DTI | Demonstration of differences in reliability with different image processing strategies. Reliability of the small-worldness index was low at low densities compared with other metrics. |
| Schwarz | 25 | <1 hour, 11±4 months | RS-fMRI | Reliability was higher over a short between-scan interval compared with a long interval. Deconvolution of white matter, CSF, and motion time series increased reliability. |
| Telesford | 45 | <1 hour | Executive task fMRI | Reliability was increased after spatial smoothing. Some metrics are more reliable in network hubs compared with nonhub nodes. Only the most stringent density thresholds significantly affected reproducibility. Mean degree was the least reliable global metric. |
| Vaessen | 6 | 14±8 days | DTI | Number of diffusion directions and gradient amplitude had no effect on reliability. |
| Wang | 25 | <1 hour, 11±4 months | RS-fMRI | Reliability was higher over a long between-scan interval compared with a short interval. Nodal metrics were more reliable than global metrics. Recommendations were given for graph construction depending on the type of atlas used. |
| Weber | 22 | 30 minutes | ASL, risk/reward task fMRI | ASL perfusion fMRI produced more reliable graph metrics than BOLD fMRI. Reliability was lowest at low frequencies. Reliability was improved during task performance. |
ASL, arterial spin-labeling perfusion functional magnetic resonance imaging; DSI, diffusion spectrum imaging; DTI, diffusion-tensor imaging; fMRI, blood oxygenation level-dependent functional magnetic resonance imaging; fNIRS, functional near-infrared spectroscopy; MEG, magnetoencephalography; ROI, regions of interest; RS, resting state.