Robert Becker^{1}, Claudia Falfan-Melgoza^{1}, Jonathan Reinwald^{1}, and Wolfgang Weber-Fahr^{1}

In a test-retest fMRI experiment we examined the influence of temporal filtering and global signal regression (GSR) on resting state networks in rats. Connectivity and topological properties as well as their test-retest reliability were assessed for eight filtering variants (with and without GSR, four frequency bands). We found GSR to strongly impair the expected structure of networks. The choice of temporal filtering frequencies whereas did not have a significant effect. Test-retest-reliability was low for all filtering variants. Based on our results we recommend to use less restrictive bandpass filters but no GSR.

Discussion

Despite generally low test-retest- reliability for the connection strength as well as global and local graph metrics, resting state networks showed the expected structure (strong DNM and SM areas). GSR was found to produce less structured and less efficient networks. The application of GSR is therefore not recommended. Regarding the choice of bandpass filter limits, the traditional choice (0.01 - 0.1 Hz) leads to networks more clearly showing the expected properties than higher frequencies. Nevertheless frequencies above 0.1 Hz yield additional connectivity information and should be take into account.1. Murphy K, Birn RM, Handwerker DA, Jones TB, Bandettini PA. The impact of global signal regression on resting state correlations: Are anti-correlated networks introduced? NeuroImage 2009; 44(3): 893-905.

2. Braun U, Plichta MM, Esslinger C, Sauer C, Haddad L, Grimm O, et al. Test-retest reliability of resting-state connectivity network characteristics using fMRI and graph theoretical measures. Neuroimage 2012; 59(2): 1404-1412.

3. Gass N, Schwarz AJ, Sartorius A, Schenker E, Risterucci C, Spedding M, et al. Sub-anesthetic ketamine modulates intrinsic BOLD connectivity within the hippocampal-prefrontal circuit in the rat. Neuropsychopharmacology 2014; 39(4): 895-906.

4. Schwarz AJ, Danckaert A, Reese T, Gozzi A, Paxinos G, Watson C, et al. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: Application to pharmacological MRI. NeuroImage 2006; 32(2): 538-550.

5. Latora V, Marchiori M. Economic small-world behavior in weighted networks. The European Physical Journal B-Condensed Matter and Complex Systems 2003; 32(2): 249-263.

6. Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010; 52(3): 1059-1069.

7. Shrout PE, Fleiss JL. Intraclass correlations: Uses in assessing rater reliability. Psychological Bulletin 1979; 86(2): 420-428.

Fig 1: Connectivity
(40% strongest edges) for non GSR (left) and GSR networks (right) and four
frequency bands each. Plots were generated using the FSL nets toolbox. Line color
indicates connection weight, line width indicates ICC (0-0.81). Clustering of
the network nodes was calculated on the network with no GSR and 0.01-0.1 Hz by
hierarchical clustering functionality implemented in fslnets. All networks were
plotted using the same clustering for comparability.

Fig 2: AUC
values (densities 0.1 – 0.4) of betweenness centrality for 46 anatomically
defined regions. Asterisks indicate significant influence (two-way ANOVA,
p<0.05, FDR corrected) of GSR (red) and choice of temporal filter (blue) respectively.

Fig 3: AUC
values (densities 0.1 – 0.4) of local graph properties degree centrality for 46
anatomically defined regions. Asterisks indicate significant influence (two-way
ANOVA, p<0.05, FDR corrected) of GSR (red) and choice of temporal filter (blue)
respectively.

Fig 4: AUC values for global network properties (densities 0.1 – 0.4).

Fig 5: Intra-class-correlation
of global network properties at densities 0.1, 0.25 and 0.4.