Longitudinal reliability of tract‐based spatial statistics in diffusion tensor imaging

Abstract

Relatively little is known about reliability of longitudinal diffusion-tensor imaging (DTI) measurements despite growing interest in using DTI to track change in white matter structure. The purpose of this study is to quantify within- and between session scan-rescan reliability of DTI-derived measures that are commonly used to describe the characteristics of neural white matter in the context of neural plasticity research. DTI data were acquired from 16 cognitively healthy older adults (mean age 68.4). We used the Tract-Based Spatial Statistics (TBSS) approach implemented in FSL, evaluating how different DTI preprocessing choices affect reliability indices. Test-Retest reliability, quantified as ICC averaged across the voxels of the TBSS skeleton, ranged from 0.524 to 0.798 depending on the specific DTI-derived measure and the applied preprocessing steps. The two main preprocessing steps that we found to improve TBSS reliability were (a) the use of a common individual template and (b) smoothing DTI data using a 1-voxel median filter. Overall our data indicate that small choices in the preprocessing pipeline have a significant effect on test-retest reliability, therefore influencing the power to detect change within a longitudinal study. Furthermore, differences in the data processing pipeline limit the comparability of results across studies.

Reliability and statistical power analysis of cortical and subcortical FreeSurfer metrics in a large sample of healthy elderly

“…As in our previous study (Madhyastha et al, 2014), we used a two-way mixed effect ICC model with absolute agreement (the ICC(A,1) model for single measures as defined by McGraw and Wong (1996)). Similar to correlations, this ICC model results in standardized values (ranging between 0 and 1); however, in contrast to correlations this metric is also sensitive to linear transformations (e.g., due to a systematic bias between time points of measurement).…”

Test–retest reliability of diffusion measures in cerebral white matter: A multiband diffusion MRI study

“…31 However, the step of skeleton extraction and projection may also bring in some biases for the estimation of diffusion measures, especially for the fiber-crossing areas. Madhyastha et al (2014) found that within the WM skeleton, there are a large percentage of voxels (> 16%) that do not correspond to each other, making it difficult to find reliable changes in these areas, which may account for the lower proportion of WM voxels with good reliability observed in TBSS analysis compared with VBA analysis.…”

“…40,41 Specifically, a recent TBSS study of WM reliability reported similar ICC values and distributions of the WM skeleton to our findings. 42 Moreover, we found that the diffusion measures from TBSS are less reliable than those from VBA.…”

Progress toward openness, transparency, and reproducibility in cognitive neuroscience

“…Researchers continue to take seriously the effort to systematically measure the factors that influence test/ retest reliability of responses across time and tasks, and these sorts of studies are increasingly common [83][84][85][86][87][88] . Other research programs focus on addressing questions about the long-term within-subject stability of responses [89][90] , and how the accurate assessment of within-participant differences might address questions about individual differences 93 .…”