Diffantom: Whole-Brain Diffusion MRI Phantoms Derived from Real Datasets of the Human Connectome Project

摘要

Diffantom in brief Diffantom is a whole-brain diffusion MRI (dMRI) phantom publicly available through the Dryad Digital Repository (doi: 10.5061/dryad.4p080 ). The dataset contains two single-shell dMRI images, along with the corresponding gradient information, packed following the BIDS standard (Brain Imaging Data Structure, Gorgolewski et al., 2015 ). The released dataset is designed for the evaluation of the impact of susceptibility distortions and benchmarking existing correction methods. In this Data Report we also release the software instruments involved in generating diffantoms , so that researchers are able to generate new phantoms derived from different subjects, and apply these data in other applications like investigating diffusion sampling schemes, the assessment of dMRI processing methods, the simulation of pathologies and imaging artifacts, etc. In summary, Diffantom is intended for unit testing of novel methods, cross-comparison of established methods, and integration testing of partial or complete processing flows to extract connectivity networks from dMRI. Introduction Fiber tracking on dMRI data has become an important tool for the in vivo investigation of the structural configuration of fiber bundles at the macroscale. Tractography is fundamental to gain information about white matter (WM) morphology in many clinical applications like neurosurgical planning (Golby et al., 2011 ), post-surgery evaluations (Toda et al., 2014 ), and the study of neurological diseases as in Chua et al. ( 2008 ) addressing multiple sclerosis and Alzheimer's disease. The analysis of structural brain networks using graph theory is also applied on tractography, for instance in the definition of the unique subject-wise patterns of connectivity (Sporns et al., 2005 ), in the assessment of neurological diseases (Griffa et al., 2013 ), and in the study of the link between structural and functional connectivity (Messé et al., 2015 ). However, the development of the field is limited by the lack of a gold standard to test and compare the wide range of methodologies available for processing and analyzing dMRI. Large efforts have been devoted to the development of physical phantoms (Lin et al., 2001 ; Campbell et al., 2005 ; Perrin et al., 2005 ; Fieremans et al., 2008 ; Tournier et al., 2008 ). C?té et al. ( 2013 ) conducted a thorough review of tractography methodologies using the so-called FiberCup phantom (Poupon et al., 2008 ; Fillard et al., 2011 ). These phantoms are appropriate to evaluate the angular resolution in fiber crossings and accuracy of direction-independent scalar parameters in very simplistic geometries. Digital simulations are increasingly popular because the complexity of whole-brain tractography can not be accounted for with current materials and proposed methodologies to build physical phantoms. Early digital phantoms started with simulation of simple geometries (Basser et al., 2000 ; G?ssl et al., 2002 ; Tournier et al., 2002 ; Leemans et al., 2005 ) to evaluate the angular resolution as well. These tools generally implemented the multi-tensor model (Alexander et al., 2001 ; Tuch et al., 2002 ) to simulate fiber crossing, fanning, kissing, etc. Close et al. ( 2009 ) presented the Numerical Fiber Generator , a software to simulate spherical shapes filled with digital fiber tracts. Caruyer et al. ( 2014 ) proposed Phantomas to simulate any kind of analytic geometry inside a sphere. Phantomas models diffusion by a restricted and a hindered compartment, similar to Assaf and Basser ( 2005 ). Wilkins et al. ( 2015 ) proposed a whole-brain simulated phantom derived from voxel-wise orientation of fibers averaged from real dMRI scans and the multi-tensor model with a compartment of isotropic diffusion. Neher et al. ( 2014 ) proposed FiberFox , a visualization software to develop complex geometries and their analytical description. Once the geometries are obtained, the software generates the corresponding dMRI signal with a methodology very close to that implemented in Phantomas . An interesting outcome of FiberFox is the phantom dataset ~(1) created for the Tractography Challenge held in ISMRM 2015. This dataset was derived from the tractography extracted in one Human Connectome Project (HCP, Van Essen et al., 2012 ) dataset. In the tractogram, 25 fiber bundles of interest were manually segmented by experts. Using FiberFox , the segmentation of each bundle was mapped to an analytical description, and finally simulated the signal. In this data report we present Diffantom , an in silico dataset to assess tractography and connectivity pipelines using dMRI real data as source microstructural information. Diffantom is inspired by the work of Wilkins et al. ( 2015 ), with two principal novelties. First, since we use a dataset from the HCP as input, data are already corrected for the most relevant distortions. The second improvement is a more advanced signal model to generate the phantom using the hindered and