Retrospective head motion correction approaches for diffusion tensor imaging: Effects of preprocessing choices on biases and reproducibility of scalar diffusion metrics

PURPOSE:To evaluate how retrospective head motion correction strategies affect the estimation of scalar metrics com-monly used in clinical diffusion tensor imaging (DTI) studies along with their across-session reproducibility errors. MATERIALS AND METHODS:Fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), axial diffusivity (AD)and their respective across-session reproducibility errors were measured on a 4T test–retest dataset of healthy partici-pants using five processing pipelines. These differed in: 1) the number of b0 volumes used for motion correction refer-ence (one or five); 2) the estimations of the gradient matrix rotation (based on 6 or 12 degrees of freedom derivedfrom coregistration); and 3) the software packages used (FSL or DTIPrep). Biases and reproducibility were evaluated inthree regions of interest (ROIs) (bilateral arcuate fasciculi, cingula, and the corpus callosum) and also at the full brainlevel with tract based skeleton images. RESULTS:Preprocessing choices affected DTI measures and their reproducibility. The DTIPrep pipeline exhibited higherDTI metrics: FA/MD and AD (P<0.05) relative to FSL pipelines both at the ROI and full brain level, and lower RD esti-mates (P<0.05) at the ROI level. Within FSL pipelines no such effects were found (P-values ranging between 0.25 and0.97). The DTIPrep pipeline showed the highest number of white matter skeleton voxels, with significantly higher repro-ducibility (P<0.001) relative to the other pipelines (tested onP<0.01 uncorrected maps). CONCLUSION:The use of an iteratively averaged b0 image as motion correction reference (as performed by DTIPrep)affects both scalar values and improves test–retest reliability relative to the other tested pipelines. These considerationsare potentially relevant for data analysis in longitudinal DTI studies.