Overview
- A peer-reviewed study published in Science Advances presents MR-AIV, a physics-informed AI framework that reconstructs 3D fluid flow from dynamic contrast MRI and time-series dye images without direct velocity measurements.
- The method trains neural networks on tracer spread and applies fluid‑physics equations to infer local flow velocity, tissue permeability, and pressure from noisy MRI data.
- Across mouse brains the team found two clearance modes: faster directed flow at brain surfaces and around blood vessels at roughly a few microns per second and much slower transport through deep tissue that is about 50 times slower.
- Key limitations are that the results are baseline animal measurements with higher uncertainty in very low-velocity regions, and the inferred pressure and permeability are model-based estimates rather than direct ground-truth readings.
- The authors say the approach could help compare healthy and diseased or aged brains and guide steps toward human-capable MRI tests for conditions such as concussion and Alzheimer’s, and the work was supported by NIH programs including the BRAIN Initiative.