Influence of diffusion pipeline on data analysis: UK Biobank example for age-diffusion dependences

摘要

Large brain data project~1 presents a great opportunity to understand different aspects of brain organisation. Combining obtained information from UK Biobank data into a common statistical framework benefits from universally adapted postprocessing pipelines for harmonised data quality assessment and manipulation. Various post-processing steps have been suggested for diffusion MRI in order to correct common sources of noise~2, Gibbs ringing correction~3, susceptibility distortion correction~4, motion correction~ 4, correction of physiological noise and outliers~5, and the influence of eddy currents to the geometrical distortions of the images~ 6. Several minimal post-processing steps have been recommended in order to prepare MRI data~(7,8). In order to assess how a chosen pipeline influences across-subject analysis we compared estimated age-curves of the diffusion metrics between pipelines using tractbased spatial statistic~9 and simple linear regression models using UK Biobank data.