A1 Refereed original research article in a scientific journal
Reproducibility and intercorrelation of graph theoretical measures in structural brain connectivity networks
Authors: Roine T, Jeurissen B, Perrone D, Aelterman J, Philips W, Sijbers J, Leemans A
Publisher: ELSEVIER SCIENCE BV
Publication year: 2019
Journal: Medical Image Analysis
Journal name in source: MEDICAL IMAGE ANALYSIS
Journal acronym: MED IMAGE ANAL
Volume: 52
First page : 56
Last page: 67
Number of pages: 12
ISSN: 1361-8415
DOI: https://doi.org/10.1016/j.media.2018.10.009
Abstract
Diffusion-weighted magnetic resonance imaging can be used to non-invasively probe the brain microstructure. In addition, recent advances have enabled the identification of complex fiber configurations present in most of the white matter. This has improved the investigation of structural connectivity with tractography methods. Whole-brain structural connectivity networks, or connectomes, are reconstructed by parcellating the gray matter and performing tractography to determine connectivity between these regions. These complex networks can be analyzed with graph theoretical methods, which measure their global and local properties. However, as these tools have only recently been applied to structural brain networks, there is little information about the reproducibility and intercorrelation of network properties, connectivity weights and fiber tractography reconstruction parameters in the brain. We studied the reproducibility and correlation in structural brain connectivity networks reconstructed with constrained spherical deconvolution based probabilistic streamlines tractography. Diffusion-weighted data from 19 subjects were acquired with b = 2800 s/mm(2) and 75 gradient orientations. Intrasubject variability was computed with residual bootstrapping. Our findings indicate that the reproducibility of graph theoretical metrics is generally excellent with the exception of betweenness centrality. A reconstruction density of approximately one million streamlines is necessary for excellent reproducibility, but the reproducibility increases further with higher densities. The reproducibility decreases, but only slightly, when switching to a higher order in constrained spherical deconvolution. Moreover, in binary networks, using sufficiently high threshold values improves the reproducibility. We show that multiple network properties and connectivity weights are highly intercorrelated. The experiments were replicated by using a test-retest dataset of 44 healthy subjects provided by the Human Connectome Project. In conclusion, our results provide guidelines for reproducible investigation of structural brain networks. (C) 2018 Published by Elsevier B.V.
Diffusion-weighted magnetic resonance imaging can be used to non-invasively probe the brain microstructure. In addition, recent advances have enabled the identification of complex fiber configurations present in most of the white matter. This has improved the investigation of structural connectivity with tractography methods. Whole-brain structural connectivity networks, or connectomes, are reconstructed by parcellating the gray matter and performing tractography to determine connectivity between these regions. These complex networks can be analyzed with graph theoretical methods, which measure their global and local properties. However, as these tools have only recently been applied to structural brain networks, there is little information about the reproducibility and intercorrelation of network properties, connectivity weights and fiber tractography reconstruction parameters in the brain. We studied the reproducibility and correlation in structural brain connectivity networks reconstructed with constrained spherical deconvolution based probabilistic streamlines tractography. Diffusion-weighted data from 19 subjects were acquired with b = 2800 s/mm(2) and 75 gradient orientations. Intrasubject variability was computed with residual bootstrapping. Our findings indicate that the reproducibility of graph theoretical metrics is generally excellent with the exception of betweenness centrality. A reconstruction density of approximately one million streamlines is necessary for excellent reproducibility, but the reproducibility increases further with higher densities. The reproducibility decreases, but only slightly, when switching to a higher order in constrained spherical deconvolution. Moreover, in binary networks, using sufficiently high threshold values improves the reproducibility. We show that multiple network properties and connectivity weights are highly intercorrelated. The experiments were replicated by using a test-retest dataset of 44 healthy subjects provided by the Human Connectome Project. In conclusion, our results provide guidelines for reproducible investigation of structural brain networks. (C) 2018 Published by Elsevier B.V.
UTU Research Portal