The Importance of Group-Wise Registration in Tract Based Spatial Statistics Study of Neurodegeneration: A Simulation Study in Alzheimer's Disease

Abstract

Tract-based spatial statistics (TBSS) is a popular method for the analysis of diffusion tensor imaging data. TBSS focuses on differences in white matter voxels with high fractional anisotropy (FA), representing the major fibre tracts, through registering all subjects to a common reference and the creation of a FA skeleton. This work considers the effect of choice of reference in the TBSS pipeline, which can be a standard template, an individual subject from the study, a study-specific template or a group-wise average. While TBSS attempts to overcome registration error by searching the neighbourhood perpendicular to the FA skeleton for the voxel with maximum FA, this projection step may not compensate for large registration errors that might occur in the presence of pathology such as atrophy in neurodegenerative diseases. This makes registration performance and choice of reference an important issue. Substantial work in the field of computational anatomy has shown the use of group-wise averages to reduce biases while avoiding the arbitrary selection of a single individual. Here, we demonstrate the impact of the choice of reference on: (a) specificity (b) sensitivity in a simulation study and (c) a real-world comparison of Alzheimer's disease patients to controls. In (a) and (b), simulated deformations and decreases in FA were applied to control subjects to simulate changes of shape and WM integrity similar to what would be seen in AD patients, in order to provide a “ground truth” for evaluating the various methods of TBSS reference. Using a group-wise average atlas as the reference outperformed other references in the TBSS pipeline in all evaluations.

Abnormal White Matter Integrity in Elderly Patients with Idiopathic Normal-Pressure Hydrocephalus: A Tract-Based Spatial Statistics Study

“…However, like any voxel-based technique, TBSS may not overcome inter-individual brain variability, especially in the presence of cerebral atrophy and ventricular enlargement observed in the elderly, AD and INPH patients. In our study, the 3 groups were matched for age and we used the study-specific template TBSS approach [13] . However, additional studies with various methods of analyzing DTI data are necessary to achieve a further improvement of DTI comparison in these groups.…”

“…First, a study-specific FA template image was generated by aligning FA images of all subjects to a common space using a nonlinear registration [13] . From this FA template image, a skeleton was created that represented the center of all white matter tracts common to all subjects.…”

White Matter Involvement in Chronic Musculoskeletal Pain

“…Data underwent eddy current correction, brain extraction 68 , and fitting with a diffusion tensor model at each voxel (DTIFIT) in order to calculate fractional anisotropy (FA), radial diffusivity (RD), and axial diffusivity (AD) maps. All raw images and processed FA maps were visually inspected for artifacts, excessive motion, and anatomical abnormalities that might interfere with registration such as lesions or enlarged cerebral ventricles due to cortical atrophy 37 . FA maps were nonlinearly registered 5,6,36 into 1×1×1mm standard space (FMRIB58_FA) and affine-aligned into MNI152 space.…”

Diffusion MRI and its Role in Neuropsychology

“…Third, detection of signal in voxels that are farther away from tract centers is reduced because they have lower contribution to the average of voxels projected to that tract location (Smith et al 2006). Fourth, when regions are located exactly between two skeleton points, they can be split into two locations when projected to nearest skeleton (Zalesky 2011)(de Groot et al 2013; Keihaninejad et al 2012)(Schwarz et al 2014) Schwarz and colleagues (Schwarz et al 2014) propose modifications to the TBSS standard pipeline that maximize (de Groot et al 2013; Keihaninejad et al 2012) through the use of groupwise registration based on Advanced Normalization Tools (ANTS; (Avants et al 2008)) or alternate non-linear registration algorithms that exclude the skeleton projection step.…”