The effects of SIFT on the reproducibility and biological accuracy of the structural connectome
Introduction
Diffusion MRI streamlines tractography is increasingly being used as one of many image analysis tools in the rapidly-evolving field of connectomics (Hagmann, 2005, Sporns et al., 2005). In this framework, the grey matter of the brain is parcellated in some manner, and the connections reconstructed using streamlines tractography used to infer structural connectivity between the parcellated areas (Hagmann et al., 2008). This allows for the evaluation of the resulting ‘connectome’ matrix using a wide range of analysis tools made possible using graph theory to make inferences about the connectional architecture of the brain (Bullmore and Sporns, 2009, Rubinov and Sporns, 2010), or perturbations to this connectivity in pathology and disease (Bassett and Bullmore, 2009, Griffa et al., 2013).
Most studies to date have used the number of streamlines connecting each node pair as a measure of ‘connection density’. This is however contrary to a fundamental limitation of streamlines tractography approaches that is well-established in the field: streamline count is not a valid marker of axonal fibre count (Jones et al., 2013). Construction of the structural connectome (or indeed any other quantitative method) using streamline count alone is therefore inadvisable. The approaches employed in the literature for addressing this issue include the following:
- •
Employing heuristics that make estimates regarding the nature of biases in the streamlines reconstruction process (e.g. increased streamline seeding in longer pathways) in order to explicitly correct for them (e.g. Hagmann et al., 2007, Colon-Perez et al., 2012). This assumes that the heuristics employed are a complete parameterization of the reconstruction biases present in the data, which is not guaranteed.
- •
Applying a threshold to generate a binary connectivity matrix in an attempt to circumvent the non-quantitative nature of streamline-based connectivity. This ‘connected-or-not-connected’ interpretation of structural connectivity may enable various sophisticated graph-theoretic analysis methods, but is not reflective of the actual underlying structure of the brain, where a wide spectrum of connectivity strength exists between various regions (Markov et al., 2011).
- •
Calculation of some quantitative parameter along the pathway connecting the nodes, rather than (or in addition to) the streamline counts (e.g. Hagmann et al., 2010, Lo et al., 2010, Pannek et al., 2013). This is comparable to using the streamlines connecting each node pair to define a mask in a voxel-based analysis, so may not be an appropriate metric for use in more complex graph-theoretic analyses due to the interpretation of the relevant quantitative parameter or the confound of crossing fibres.
We recently proposed the “Spherical-deconvolution Informed Filtering of Tractograms (SIFT)” algorithm as a mechanism for addressing these quantification issues (Smith et al., 2013). By imposing a model that maps a streamlines reconstruction back to the acquired diffusion image data, and modifying a reconstruction to improve its correspondence with the image data given this model, the number of streamlines connecting two regions of the brain becomes a proportional estimate of the total cross-sectional area of the white matter fibre pathway connecting those regions; this is inherently a highly biologically relevant measure of ‘structural connectivity’. We therefore advocate that such a processing step is essential to ensure that any conclusions drawn about the structural connectedness of the brain, or differences in structural connectivity between subjects, are biologically relevant and not due to systematic errors in reconstruction processes and analyses.
In this work, we interrogated the effects of the SIFT algorithm on the estimated structural connectome in three ways. Firstly, we use a simple synthetic phantom to highlight the importance of imposing such a model on a streamlines reconstruction before the connectome is generated. Secondly, we investigated the effects of SIFT on the reproducibility of the structural connectome. Thirdly, we compared the tractograms and connectomes with and without the application of SIFT to quantitative and qualitative estimates of white matter connectivity derived from published post mortem brain dissection experiments; given the lack of a true quantitative gold standard to assess whole-brain tracking results, these ex vivo results provide important (if imperfect) measures of white matter connectivity from a source other than diffusion MRI to which streamlines reconstructions may be compared.
Section snippets
Implications of the SIFT model
In the SIFT method, a simple model is imposed that maps a streamlines reconstruction back to the measured diffusion signal, as a means for reducing some of the major reconstruction biases of streamlines tractography. The model is described in detail in Smith et al. (2013), and is summarized briefly below:
- •
The density of a discrete fibre population within any voxel in the image can be estimated (up to a global scaling factor) using the integral of the relevant lobe of the Fibre Orientation
Results
Fig. 3 shows examples of the four types of streamlines reconstructions, and the resulting connectome matrices, using data from the intra-scan case. For visualization purpose, the streamlines reconstructions are displayed using Track Density Imaging (TDI) (Calamante et al., 2010), which allows for visualisation of streamlines density throughout the example slice, providing a succinct visual representation of the differences introduced by varying the type of reconstruction in the relative
Discussion
We have previously proposed the SIFT method as a mechanism for reducing some of the reconstruction biases associated with diffusion MRI streamlines tractography, by imposing physical volumetric constraints that are otherwise ignored by the streamlines reconstruction process (Smith et al., 2013). By doing so, each reconstruction should more closely mimic the underlying biology by design. In this study, we have provided three principal sources of evidence to support the hypothesis that
Conclusion
In order for the analysis of a brain structural connectome to be valid and its results to be interpretable, it is essential that the derivation of the underlying connection densities between brain regions be robust and biologically meaningful. We have shown that due to the physical underpinnings of the SIFT method, its application results in estimates of structural connectivity that are both more reproducible, and more representative of the underlying biological white matter connectivity. This
Acknowledgments
We are grateful to the National Health and Medical Research Council (NHMRC) of Australia, the Australian Research Council, Austin Health, and the Victorian Government's Operational Infrastructure Support Program for their support. This research was supported by a Victorian Life Sciences Computation Initiative (VLSCI) grant [VR0271] on its Peak Computing Facility at the University of Melbourne, an initiative of the Victorian Government, Australia.
References (54)
- et al.
Fiber composition of the human corpus callosum
Brain Res.
(1992) - et al.
How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging
NeuroImage
(2003) - et al.
Conserved and variable architecture of human white matter connectivity
NeuroImage
(2011) - et al.
Human cortical connectome reconstruction from diffusion weighted MRI: the effect of tractography algorithm
NeuroImage
(2012) - et al.
Track-density imaging (TDI): super-resolution white matter imaging using whole-brain track-density mapping
NeuroImage
(2010) - et al.
Mapping the human connectome at multiple scales with diffusion spectrum MRI
J. Neurosci. Methods
(2012) - et al.
Characteristics and variability of structural networks derived from diffusion tensor imaging
NeuroImage
(2012) - et al.
Wiring optimization in cortical circuits
Neuron
(2002) - et al.
Cortical surface-based analysis: I. Segmentation and surface reconstruction
NeuroImage
(1999) - et al.
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
NeuroImage
(2006)
Structural connectomics in brain diseases
NeuroImage
History of neuromorphometry
J. Neurosci. Methods
k-Space and q-space: combining ultra-high spatial and angular resolution in diffusion imaging using ZOOPPA at 7 T
NeuroImage
White matter integrity, fiber count, and other fallacies: the do's and don'ts of diffusion MRI
NeuroImage
A Bayesian model of shape and appearance for subcortical brain segmentation
NeuroImage
Apparent fibre density: a novel measure for the analysis of diffusion-weighted magnetic resonance images
NeuroImage
Complex network measures of brain connectivity: uses and interpretations
NeuroImage
Advances in functional and structural MR image analysis and implementation as FSL
NeuroImage
Anatomically-constrained tractography: improved diffusion MRI streamlines tractography through effective use of anatomical information
NeuroImage
SIFT: spherical-deconvolution informed filtering of tractograms
NeuroImage
Advances in diffusion MRI acquisition and processing in the Human Connectome Project
NeuroImage
Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution
NeuroImage
Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution
NeuroImage
The effect and reproducibility of different clinical DTI gradient sets on small world brain connectivity measures
NeuroImage
Whole-brain anatomical networks: does the choice of nodes matter?
NeuroImage
Human brain networks in health and disease
Curr. Opin. Neurol.
Efficient physical embedding of topologically complex information processing networks in brains and computer circuits
PLoS Comput. Biol.
Cited by (0)
- 1
Current address: Centre for the Developing Brain, King's College London, London, United Kingdom.
- 2
Current address: Department of Biomedical Engineering, Division of Imaging Sciences & Biomedical Engineering, King's College London, London, United Kingdom.