Statistical Instability of TBSS Analysis Based on DTI Fitting Algorithm

Abstract

Voxel-based DTI analysis is an important approach in the comparison of subject groups by detecting and localizing gray and white matter changes in the brain. One of the principal problems for intersubject comparison is the absence of a "gold standard" processing pipeline. As a result, contradictory results may be obtained from identical data using different data processing pipelines, for example, in the data normalization or smoothing procedures. Tract-based spatial statistics (TBSS) shows potential to overcome this problem by automatic detection of white matter changes and decreasing variation in the performed analysis. However, skeleton projection approaches, such as TBSS, critically depend on the accuracy of the diffusion scalar metric estimations. In this work, we demonstrate that the agreement and reliability of TBSS results depend on the applied DTI data processing algorithm. Statistical tests have been performed using two in vivo measured datasets and compared with different implementations of the least squares algorithm. As a result, we recommend repeating TBSS analysis using different fitting algorithms, in particular, using on iteratively-assessed robust estimators, as accurate and more reliable approach in voxel-based analysis, particularly, for TBSS. Repeating TBSS analysis allows one to detect and localize suspicious regions in white matter which were estimated as the regions with significant difference. Finally, we did not find a favorite fitting algorithm (or class of them) which can be marked as more reliable for group comparison.

Feasibility of axon density metrics for brain asymmetry evaluation in the UK Biobank subsample

“…3b). It emphasises the need in accurately harmonised data before analysis, in particular, in the noise correction as an important step in post-processing pipeline and carefully to consider a numerical implementation of the used models (David et al, 2019), (Maximov et al, 2015).…”

Towards an optimised processing pipeline for diffusion magnetic resonance imaging data: Effects of artefact corrections on diffusion metrics and their age associations in UK Biobank

“…Various postprocessing steps have been suggested to correct specific sources of noise and distortions, including thermal noise evaluation (Veraart, Novikov, et al, ; Veraart, Fieremans, & Novikov, ), Gibbs ringing correction (Kellner, Dhital, Kiselev, & Reisert, ; Veraart, Fieremans, Jelescu, Knoll, & Novikov, ), susceptibility distortion correction (Andersson & Sotiropoulos, ), motion correction (Andersson, Graham, Zsoldos, & Sotiropoulos, ; Andersson & Sotiropoulos, ), correction of physiological noise and outliers (Maximov et al, ; Maximov, Grinberg, & Shah, ; Sairanen, Leemans, & Tax, ; Walker et al, ) and eddy current induced geometrical distortions (Taylor et al, ). However, although the application of even part of the postprocessing steps such as noise correction has been demonstrated to improve sensitivity and provide additional information about absolute diffusion metrics (Kochunov et al, ), systematic evaluations of the effects of the different steps on the diffusion metrics are scarce.…”

Fast qualitYconTrol meThod foRderIved diffUsionMetrics (YTTRIUM) in big data analysis: UK Biobank 18608 example

“…Various approaches have been developed to detect and correct artefacts in raw diffusion data originating, e.g. from eddy currents, bulk head motions, susceptibility distortions (Andersson and Sotiropoulos 2016), noise (Kochunov et al 2018), presence of outliers (Koch et al 2019), and diffusion metric variability (Maximov et al 2015), (David et al 2019).…”

“…However, QC and data harmonisation procedures applied on raw diffusion data (Mirzaalian et al 2018), (Fortin et al 2017) do not guarantee accurate numerical computation of scalar diffusion metrics. Derived diffusion metrics from diffusion or kurtosis tensors are sensitive to a range of subjectspecific factors such as age or various brain disorders, but also to applied numerical algorithm or its programming implementation (Lebel et al 2012), (Grinberg et al 2017), (Maximov et al 2015), (David et al 2019). The effects of noisy observations on subsequent between-subjects analysis involving the derived diffusion metrics can be mitigated using simple outlier detection procedures (see, for example, (Richard et al 2018), (Tønnesen et al 2018), (de Lange et al 2019)).…”