Eirini Messaritaki^{1}, Stavros I Dimitriadis^{1,2}, and Derek K Jones^{2,3}

Structural brain networks derived from diffusion Magnetic Resonance Imaging data can use various tract metrics to weigh the network edges. In this work we use the Human Connectome Project test-retest diffusion MRI data to assess the reproducibility of structural brain networks, their edges and their graph-theoretical measures derived using different edge-weighting strategies.

Using the NS as edge weights
resulted in the highest reproducibility (comparable to ^{7,8}). Combining NS and FA (denoted 'NS+FA') to form integrated
graphs resulted in the second highest reproducibility (Figure 1). The
reproducibility of the FA- and MD-weighted graphs was very low. Interestingly, the
reproducibility of the NS+FA integrated graphs was negatively correlated (r =
-0.45, *p* =0.0048) with the time interval between test and retest scans. Further analysis focused on the graphs that use as edge weights a) only the NS, b) the NS+FA
combination. In the NS-graphs, there were 16 edges
common across both scans of all participants, while in the NS+FA integrated graphs
there were 28 such edges.

A
representative graph for the structural network of one participant is shown in Figure 2, where (a) shows the NS-graph and (b) shows the NS+FA integrated graph. The ICC for the edges that
appear in all participants is shown in Figure 3 (blue circles for the NS-graph and red squares for the NS+FA integrated graph). The ICCs were comparable for the NS and the NS+FA
networks. Additionally, for both the NS-graphs and the NS+FA integrated graphs, the edge ICC was
statistically significantly correlated with the mean number (over participants
and over the two scans) of streamlines of the edge (*p*-values of 10^{-7} and 10^{-28}, and correlation
coefficients of 0.28 and 0.39, for the NS- and the (NS+FA) graphs respectively).
The ICC for the global
efficiency of the graphs was 0.77 for the NS-networks and 0.54 for the NS+FA
networks. The lower value of the ICC in the latter case was driven by larger
differences in participants for whom the scans differ by more than 6 months.

[1] D. C. Van Essen, S. M. Smith, D. M. Barch, T. E. J. Behrens, E. Yacoub, K. Ugurbil, for the WU-Minn HCP Consortium. The WU-Minn Human Connectome Project: An overview. NeuroImage, 80:62-79, 2013.

[2] M. F. Glasser, S. N.Sotiropoulos, J. A. Wilson, T. S.Coalson, B. Fischld, J. L. Andersson, J. Xu, S. Jbabdi, M. Webster, J. R. Polimeni, D. C. Van Essen, M. Jenkinson for the WU-Minn HCP Consortium. The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage, 80:105-124, 2013.

[3] A. Leemans, B. Jeurissen, J. Sijbers, and D.K. Jones. ExploreDTI: a graphical toolbox for processing, analyzing, and visualizing diffusion MR data. 17th Annual Meeting of Intl Soc Mag Reson Med, p. 3537, Hawaii, USA, 2009.

[4] N. Tzourio-Mazoyer, B. Landeau, D. Papathanassiou, F. Crivello, O. Étard, N. Delcroix, B. Mazoyer, and M. Joliot. Automated Anatomical Labeling of Activations in SPM Using a Macroscopic Anatomical Parcellation of the MNI MRI Single-Subject Brain. NeuroImage, 15 :273-289, 2002.

[5] S. I. Dimitriadis, M. Drakesmith, S. Bells, G. D. Parker, D. E. Linden and D. K. Jones. Improving the Reliability of Network Metrics in Structural Brain Networks by Integrating Different Network Weighting Strategies into a Single Graph. Frontiers in Neuroscience, 11:694, 2017.

[6] M. Rubinov and O. Sporns. Complex network measures of brain connectivity: Uses and interpretations. NeuroImage 52:1059-69, 2010.

[7] T. Roine, B. Jeurissen, D. Perrone, J. Aelterman, W. Philips, J. Sijbers, A. Leemans. Reproducibility and intercorrelation of graph theoretical measures in structural brain connectivity networks. Medical Image Analysis. article in press.

[8] S.-Y. Tsai. Reproducibility of structural brain connectivity and network metrics using probabilistic diffusion tractography. Scientific Reports, Volume 8, Article number: 11562 (2018).

Figure 1:
Reproducibility of structural networks, when different WM tract metrics (or
their combinations) are used as edge weights, versus the time interval between
test and retest. Using NS as the edge weights results in the highest
reproducibility. Using a linear combination of NS and FA as edge
weights results in high reproducibility, which also exhibits negative
correlation with the time between test and retest. Using FA or MD on their own
as edge weights results in very low reproducibility.

Figure 2: Representative connectome for one participant using as graph edge weights (a)
the NS, and (b) the NS+FA combination. The NS+FA integrated connectome
exhibits connections that are not present in the one on the left.

Figure 3: Edge ICCs for the edges that are present in all 37 participants.
Blue circles: NS-graphs. Red squares: NS+FA graphs. The range of edge ICCs is comparable for both edge-weighting strategies.