# Physics-Informed AI Maps Glymphatic Fluid Velocity > Source: > Published: 2026-05-27 20:48:33.571370+00:00 # Physics-Informed AI Maps Glymphatic Fluid Velocity A multidisciplinary team led by Professor Douglas Kelley at the University of Rochester used physics-informed neural networks to infer brain-wide glymphatic fluid velocities from MRI tracer videos, according to a University of Rochester press release and coverage in NeuroscienceNews and MedicalXpress. The method, which the authors call MR-AIV in the paper available on PMC, trains constraints from known physics on neural networks to recover both local fluid velocity and tissue permeability from slow-moving tracer dispersion (University of Rochester; PMC). Results in the reported study show a dual-speed drainage pattern: a "fast track" flowing at a few microns per second across cortical surfaces and a "slow track" in deep tissue roughly **50x** slower (University of Rochester; NeuroscienceNews). The work was published in Science Advances and has been demonstrated in animal models while researchers work on adapting the pipeline for human MRI studies (University of Rochester; NeuroscienceNews). ### What happened A new study published in **Science Advances**, reported by the University of Rochester on May 27, 2026, uses physics-informed artificial intelligence to infer brain-wide glymphatic fluid velocities from magnetic resonance imaging tracer videos (University of Rochester press release; MedicalXpress). The authors name their framework MR-AIV in the paper available via PMC, and they report that the trained networks recover both local fluid flow velocity and tissue permeability from the temporal dispersion of dye tracers captured by MRI (PMC; University of Rochester). ### Technical details The published approach applies **physics-informed neural networks** that embed conservation laws and transport physics as constraints during training, allowing inference of extremely slow flows that conventional MRI velocity-encoding cannot resolve (PMC; NeuroscienceNews). The team trained models on time-series MRI of tracer dye propagation in preclinical brains and used the learned representation to map velocity fields and spatially varying permeability. Per the University of Rochester reporting, the maps reveal two distinct clearance regimes: a "fast track" moving at **a few microns per second** across open cortical regions and a "slow track" within deep tissue that is about **50x** slower (University of Rochester; NeuroscienceNews). ### Editorial analysis - technical context Physics-informed machine learning is increasingly used to infer latent physical fields from noisy, undersampled measurements; this study is a domain-specific application that leverages those strengths for neuroimaging. For practitioners, the key technical ingredients are: - •incorporating physical PDE constraints to regularize inverse problems - •using tracer time-series as the observable rather than relying on direct velocity encoding - •jointly estimating permeability and velocity to separate tissue transport properties from bulk flow. These patterns align with recent MR-AI velocimetry work in other fluid systems (PMC preprint; bioRxiv snippet) ### Context and significance Mapping glymphatic flow rates at whole-brain scale addresses a long-standing measurement gap in neuroscience because traditional microscopy is spatially local and standard MRI cannot directly measure the millimeter-to-micron-per-second flows involved. The reported **dual-speed** architecture links surface-proximal rapid clearance and much slower parenchymal transport, which is relevant to hypotheses about how metabolic waste such as amyloid-beta is distributed and removed in sleep and disease contexts (University of Rochester; NeuroscienceNews). For ML practitioners, the study is a concrete example where physics-informed models turn temporally resolved but indirect measurements into interpretable physical fields useful for biology and translational research. ### What to watch - •Validation in humans: public coverage notes the team is adapting the pipeline for human MRI and plans comparisons across age and disease cohorts; follow replication and sensitivity analyses in clinical scans (University of Rochester; NeuroscienceNews). - •Robustness and generalization: assess how model estimates depend on tracer properties, MRI temporal resolution, and noise; independent reproduction and open code or model checkpoints will be important (PMC; MedicalXpress). - •Biomarker development: observers will watch whether inferred velocity/permeability metrics correlate with established clinical measures or cognitive outcomes, enabling use in trials or longitudinal studies (NeuroscienceNews). ### Takeaway This study exemplifies how embedding domain physics in neural networks can unlock latent variables from standard imaging modalities. The reported MR-AIV framework and its dual-speed glymphatic maps form a testable set of biomarkers in preclinical models; the next critical steps reported in media coverage are human adaptation, cross-site replication, and evaluation against clinical endpoints (University of Rochester; NeuroscienceNews; PMC). ## Scoring Rationale The paper applies physics-informed ML to a previously intractable measurement, producing interpretable velocity maps with potential translational value. It is notable for computational neuroimaging and inverse-problem practitioners, though current demonstrations are preclinical and require human validation. Practice with real Ad Tech data 90 SQL & Python problems · 15 industry datasets [Active Search Campaigns by BudgetEasy](/problems/sql/active-search-campaigns-by-budget) [High CPC Clicks & Poor Landing PagesMedium](/problems/sql/high-cpc-clicks-poor-landing-page) [Campaign ROAS by Attribution ModelHard](/problems/sql/campaign-roas-by-attribution-model) 250 free problems · No credit card [See all Ad Tech problems](/problems/datasets/adtech)