Data-driven microstructure imaging with multi-dimensional diffusion MRI in early brain development
Understanding how human brain structure develops and grows before and around the time of birth is a grand challenge in neuroscience, with far reaching implications for our understanding of neuropsychiatric disorders. Diffusion MRI (dMRI) and other quantitative MRI relaxometry modalities offer a unique means to characterize neural tissue microstructure in vivo. However, analysis of multi-dimensional dMRI in neonatal and foetal data brings about specific challenges.
First, current MRI microstructure imaging is based on models that are largely designed for mature brain tissue and ill adapted to the rapidly changing microstructure in early brain development. We therefore envisage a generative approach, in which the tissue microstructure is learned from and hence adapted to the data at hand, using unsupervised machine learning methodology with as few model assumptions as possible.
Second, analysis of infant MRI is complicated by unavoidable head motion during acquisition, resulting in a collection of scattered image slices that need to be reconstructed in an anatomical reference frame. Building on our experience with dMRI motion correction for foetal and neonatal imaging, we aim to develop the first integrated slice-to-volume reconstruction for multi-dimensional dMRI.
When combined, these developments will open new avenues for exploring and quantifying microstructural changes and facilitate longitudinal group studies of normal and abnormal brain development.