We get some nice looking bundles too, even in weeks-old brains:

We checked that these microstructure maps (controlling for qc scores and site) replicate developmental findings: FA and RTOP increase with age, MD decreases. NDC relates strongly to FA/MD

We split the dMRI postprocessing workflows (model fitting, tractography, parametric microstructure map calculation) into QSIRecon, a containerized BIDS App that runs on the outputs from QSIPrep. QSIRecon delivers >40 parametric maps, personalized bundles and bundle metrics per session for HBCD.

Other effects aren’t obvious in individual images, but their effects are clear in group maps. Eg., using DRBUDDI after TOPUP/eddy decreases misalignment of high-FA structures. Using MP-PCA and Gibbs unringing increases qc measures like Contrast to Noise Ratio and Neighboring DWI Correlation.

The effects of some image correction steps are easy to see: correction of susceptibility distortion, signal pileup and slice dropout are mostly mitigated

We made numerous updates to the QSIPrep and QSIRecon software packages to better perform on HBCD’s infant/pediatric diffusion MRI. New preprocessing includes partial-Fourier Gibbs unringing, MP-PCA, 2-stage susceptibility distortion correction, SynthStrip/Synthseg and MNIInfant atlases.