Data-Efficient Neural Operators Enable Brain-Wide Glymphatic Velocity Field Estimation from DCE-MRI

Estimating cerebrospinal fluid (CSF) flow speeds in the brain is essential for understanding solute transport in the glymphatic system. Because brainwide flow velocities cannot be directly measured in vivo, they must be inferred from indirect imaging data. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provides spatiotemporal concentration measurements, but reconstructing full three-dimensional velocity fields, which span several orders of magnitude, from noisy data remains a challenging ill-posed inverse problem. In this work, we present a DCE-MRI-based neural operator framework for estimating CSF velocity fields from concentration time series. The method leverages proper orthogonal decomposition (POD) of DCE-MRI data to obtain compact spatial-temporal representations that are subsequently used by neural operators to reconstruct three-dimensional flow fields. By decomposing the brain volume along orthogonal planes and retaining dominant POD modes, the neural operators improve robustness to noise and reduce the dimensionality of the reconstruction problem. The framework is first evaluated using synthetic DCE-MRI datasets generated from 45 anatomically realistic mouse brain models with varying geometries and permeability fields. Quantitative comparisons demonstrate improved velocity reconstruction accuracy relative to baseline deep learning reconstructions, particularly in regions with elevated flow speeds. Additional experiments using permeability fields derived from real mouse brains show that accurate velocity estimation can be achieved with limited training data, indicating strong data efficiency and generalizability. Finally, we demonstrate a simulation-to-measurement transfer strategy in which models trained on synthetic DCE-MRI data are refined using a small number of realistic measurements, yielding physically plausible brain-wide CSF velocity estimates. These results highlight the potential of DCE-MRI-based reconstruction frameworks for noninvasive imaging of glymphatic transport