Reproducibility of small world metrics from diffusion tensor tractography at 3 Tesla M.J. Vaessen1,2, P.A. Hofman1,2, J.F. Jansen1, H.N. Tijssen1,3, A. Aldenkamp2, W.H. Backes1 Department of Radiology, Maastricht University Medical Centre, Maastricht, Netherlands, 2Epilepsy Centre Kempenhaeghe, Heeze, Netherlands, 3Centre for functional MRI of the Brain (FMRIB), University of Oxford, Oxford, UK
1
L
K
Introduction Advances in computational network analysis have enabled the characterization of topological properties in large scale networks such as the human brain. Recent studies demonstrated that structural brain networks have small-world attributes [1-3]. Small-world networks are characterized by a topology in which most nodes are not neighbors of one another, but can be reached through a small number of steps. Their attributes may serve as clinical markers in a variety of pathologies [4]. However, little is known about their accuracy and reproducibility, which have consequences for detectable clinical differences. Moreover, acquisition parameters, such as number of gradient directions and diffusion strength, possibly influence network metrics derived from tractography data. The aim of the present study is twofold: (i) to determine the effect of clinically available DTI sampling schemes on small world metrics derived from tractography data and (ii) to asses the test-retest reproducibility of small world metrics. Materials and methods In vivo experiments. Single-shot spin-echo echo-planar-imaging (SE-EPI) DTI experiments were conducted on six healthy volunteers (5 male, 1 female, aged 45 23-28 years) on a 3T Philips Achieva whole body scanner (Philips Medical Systems, Best, the Netherlands). Each subject was scanned twice (average interval 14±8 days). Six sampling schemes, which varied in directional 40 resolution (Ndir = 32, 15 and 6) and gradient amplitude (Gamp) to obtain shorter echo time (TE, 66 ms and 56 ms), were employed in random order. 35 Sampling schemes were matched for total scan time (12 min) by setting number 6+ 6 15+ 15 32+ 32 of averages appropriately. Anatomic reference images were acquired by a T1 1.65 weighted 3D-TSE sequence. Image analysis. Whole brain connectivity data was obtained by tracking from Brodmann areas, defined on the grey-white matter interface, using probabilistic 1.575 tractography (PICo [5]) (Fig. 1). a
1.5
b
6+
6
15+ 15 32+ 32
6+
6 15+ 15 32+ 32 Sampling scheme
C
0.75 0.72 0.69
Figure 1: (a) Example of the grey-white matter interface. (b) Resulting connections derived from the tractography data.
Figure 2: Mean ± SD of small world metrics (‘+’ indicates higher Gamp).
# tracts
32+ 6000 From the connectivity data the following small world network metrics were 32 computed: average node degree (K), characteristic path length (L), and cluster coefficient (C). Effect of gradient sampling scheme was analyzed using a two15+ 4000 way ANOVA test with a Tukey HSD post hoc test. The accuracy was expressed 15 in terms of two reproducibility measures: the Coefficient of Variation (CV) and 6+ Reproducibility Coefficient (RC). 6 Results 2000 The values of the small-world metrics for each sampling scheme are shown in Fig 2. Node degree K and cluster coefficient C increased significantly (p<0.05) with directional resolution. Characteristic path length L showed significant 0 0 50 100 decrease (p<0.05) with directional resolution. Lower directional resolution is associated with less long range tracts (Fig. 3). Gradient amplitude did not have a Tract length (mm) significant effect on any of the metrics. There was no significant deviation in Figure 3: Histogram of tract length. CV values for either Ndir or Gamp. RC values also showed no significant differences for Ndir and Gamp. Discussion Small world metrics derived from tractography data are influenced by the employed gradient scheme. Less long range tracts (Fig. 3), imply that distant brain regions are less connected, this could explain differences in small world metrics. The higher K and C, and lower L values observed in the schemes with higher directional resolution reflect more efficient networks, which are likely to be closer to the expected values for structural networks of the healthy brain. Small world metrics show low CV and RC (Table 1), advocating the applicability of these measures in clinical studies to detect small effects. References K L C [1] D. S. Bassett et al., Neuroscientist 12, 512 (2006). [2] O. Sporns, et al., Mean 43.760 1.528 0.735 Neuroinformatics 2, 145 (2004). [3] P. Hagmann et al., PLoS Biol 6, e159 SD 1.277 0.015 0.008 (2008). [4] Y. Liu et al., Brain 131, 945 (2008). [5] G. J. Parker et al., JMRI CV 0.021 0.007 0.009 18, 242 (2003). RC 2.502 0.029 0.017
Table 1: Reproducibility values for the ‘32’ sampling scheme.