mrHARDIflow : A pipeline tailored for the preprocessing and analysis of Multi-Resolution High Angular diffusion MRI and its application to a variability study of the PRIME-DE database

The validation of advanced methods in diffusion MRI requires finer acquisition resolutions, which is hard to acquire with decent Signal-to-Noise Ratio (SNR) in humans. The use of Non-Human Primates (NHP) and anaesthesia is key to unlock valid microstructural maps, but tools must be adapted and configured finely for them to work well. Here, we propose a novel processing pipeline implemented in Nextflow, designed for robustness and scalability, in a modular fashion to allow for maintainability and a high level of customization and parametrization, tailored for the analysis of diffusion data acquired on multiple spatial resolutions. Modules of processes and workflows were implemented upon cutting edge and state-of-the-art MRI processing technologies and diffusion modelling algorithms, namely Diffusion Tensor Imaging (DTI), Constrained Spherical Deconvolution (CSD) and DIstribution of Anisotropic MicrOstructural eNvironments in Diffusion-compartment imaging (DIAMOND), a multi-tensor distribution estimator. Using our pipeline, we provide an in-depth study of the variability of diffusion models and measurements computed on 32 subjects from 3 sites of the PRIME-DE, a database containing anatomical (T1, T2), functional (fMRI) and diffusion (DWI) imaging of Non-Human Primate (NHP). Together, they offer images acquired over a range of different spatial resolutions, using single-shell and multi-shell b-value gradient samplings, on multiple scanner vendors, that present artifacts at different level of importance. We also perform a reproducibility study of DTI, CSD and DIAMOND measurements outputed by the pipeline, using the Aix-Marseilles site, to ensure our implementation has minimal impact on their variability. We observe very high reproducibility from a majority of diffusion measurements, only gamma distribution parameters computed on the DIAMOND model display a less reproducible behaviour. This should be taken into consideration when future applications are performed. We also show that even if promising, the PRIME-DE diffusion data exhibits a great level of variability and its usage should be done with care to prevent instilling uncertainty in statistical analyses.