Kernel regression estimation of fiber orientation mixtures in diffusion MRI

Abstract

We present and evaluate a method for kernel regression estimation of fiber orientations and associated volume fractions for diffusion MR tractography and population-based atlas construction in clinical imaging studies of brain white matter. This is a model-based image processing technique in which representative fiber models are estimated from collections of component fiber models in model-valued image data. This extends prior work in nonparametric image processing and multi-compartment processing to provide computational tools for image interpolation, smoothing, and fusion with fiber orientation mixtures. In contrast to related work on multi-compartment processing, this approach is based on directional measures of divergence and includes data-adaptive extensions for model selection and bilateral filtering. This is useful for reconstructing complex anatomical features in clinical datasets analyzed with the ball-and-sticks model, and our framework’s data-adaptive extensions are potentially useful for general multi-compartment image processing. We experimentally evaluate our approach with both synthetic data from computational phantoms and in vivo clinical data from human subjects. With synthetic data experiments, we evaluate performance based on errors in fiber orientation, volume fraction, compartment count, and tractography-based connectivity. With in vivo data experiments, we first show improved scan-rescan reproducibility and reliability of quantitative fiber bundle metrics, including mean length, volume, streamline count, and mean volume fraction. We then demonstrate the creation of a multi-fiber tractography atlas from a population of 80 human subjects. In comparison to single tensor atlasing, our multi-fiber atlas shows more complete features of known fiber bundles and includes reconstructions of the lateral projections of the corpus callosum and complex fronto-parietal connections of the superior longitudinal fasciculus I, II, and III.

Investigating white matter fibre density and morphology using fixel-based analysis

“…Cabeen et al, 2015; Jbabdi et al, 2010; Panagiotaki et al, 2012; Scherrer and Warfield, 2012). This is a more exact description, and in the context of modulation by FC, it makes more sense to preserve the sum of the ‘fibre volume fractions’ across a bundle's width compared to preserving the sum of the ‘fibre density’ (Fig.…”

“…population fraction of the restricted compartment (Assaf and Basser, 2005), restricted fraction (De Santis et al, 2014a, De Santis et al, 2014b), axonal density (Assaf et al, 2008, De Santis et al, 2014a, De Santis et al, 2014b, Dyrby et al, 2013), partial volume fraction (Jbabdi et al, 2010), fibre density (Alexander et al, 2010, Assaf et al, 2013, Reisert et al, 2013, Riffert et al, 2014), apparent fibre density (Dell’acqua et al, 2010, Raffelt et al, 2012b), neurite density (Jespersen et al, 2010, Zhang et al, 2012), intra-axonal volume fraction (Panagiotaki et al, 2012) fibre volume fraction (Cabeen et al, 2015), fascicle fraction of occupancy (Scherrer et al, 2016)). While there are advantages and disadvantages to the different terminologies, in this work we refer to it as “fibre density” (FD) (see Section 5 for further comment on nomenclature).…”

A Comparative evaluation of voxel-based spatial mapping in diffusion tensor imaging

“…predominantly single fiber regions in deep white matter [64]. This could be more thoroughly studied by examining reliability of multi-fiber extensions of TBSS [65], possibly with multi-compartment model smoothing [66]. Voxel-based analysis had low reproducibility in superficial and periventric-ular white matter, with CV above 7% and ICC below 0.5 in some cases; however, region-based analysis was found to have lower spatial variability and better performance in these areas.…”

In vivo Exploration of the Connectivity between the Subthalamic Nucleus and the Globus Pallidus in the Human Brain Using Multi-Fiber Tractography

“…The fiber tracking algorithm used a step size of 0.5 mm and 25 seeds per voxel, which were placed in a one-voxel neighborhood surrounding each ROI. The ball-and-stick diffusion models were interpolated using a data-adaptative kernel regression framework (Cabeen et al, 2013, 2016). The framework used clustering-based optimization and two parameters: spatial bandwidth, which controls the smoothness of the interpolation, and model selection, which controls the number of estimated compartments.…”