The technical breakthrough: teaching AI the laws of fluid dynamics

Conventional MRI can show that a tracer dye is spreading through the brain, but it cannot measure the velocity of that spread when movement occurs at just a few microns per second . To overcome this, the researchers built neural networks and trained them on time-lapse MRI videos of dye dispersing through brain tissue. The key innovation was feeding the AI not just the imaging data, but also the fundamental equations of fluid dynamics .

This approach, known as physics-informed neural networks or artificial intelligence velocimetry (AIV), forces the model’s predictions to obey the laws of physics . By doing so, the AI could infer two previously inaccessible parameters from the dye’s movement: the local velocity of the fluid and the permeability of the surrounding brain tissue . The technique builds on earlier research by the same group that used AIV to quantify pressure, wall shear stress, and 3D velocities in murine perivascular spaces .

The discovery: a dual-speed clearance system

The AI-powered reconstruction exposed a stark contrast in how the glymphatic system moves fluid depending on location :

• Fast flow at the surface: In open regions, such as the subarachnoid space between the skull and the brain, CSF moves at a pace of a few microns per second, efficiently sweeping along major arterial highways.
• Slow flow deep in tissue: Once fluid penetrates the brain’s dense parenchyma (the functional tissue), it trickles through at a rate roughly 50 times slower than surface flow .

This dual-speed regime makes biological sense. The brain’s outer surface acts as a high-conductance distribution network, while deep tissue presents high hydraulic resistance, causing fluid to percolate slowly through narrow interstitial spaces . Previous modeling work by Kelley’s group had already pointed toward low-resistance perivascular spaces coupled with high-resistance parenchyma as the only configuration that could drive glymphatic flow with a small pressure drop and allow good perfusion throughout the cortex . The new AI measurements now provide direct in vivo evidence for this structure.

Beyond velocity: measuring tissue permeability

A major hidden variable in glymphatic research has been tissue permeability—how easily brain tissue allows fluid to pass through it. The new physics-informed AI framework simultaneously infers permeability by observing how the tracer disperses and constraining the solution with conservation laws . Changes in brain tissue permeability could be an early marker of pathology; if tissue becomes more resistant to fluid flow, waste clearance stalls. Being able to measure this property non-invasively from MRI could open a new window into the earliest stages of neurodegenerative disease.

Current limitations: the gap between mice and humans

It is critical to note that all measurements so far have been made in animal models, specifically mice, to establish baseline values . Human brain imaging presents significant additional hurdles, including larger scale, longer scan times, and the need for clinically safe tracer agents. The researchers are actively working toward adapting the method for human application, but this translational step remains a work in progress .

Clinical possibilities on the horizon

Even with these caveats, the long-term clinical possibilities are striking. The ability to directly measure glymphatic function from a standard MRI scan could one day transform neurology:

• Alzheimer’s disease screening: The glymphatic system is responsible for clearing amyloid beta proteins during sleep . If the AI method can be translated to humans, it could potentially measure whether an Alzheimer’s patient has impaired brain fluid circulation, or even screen healthy individuals earlier in life for poor clearance before plaques accumulate .
• Traumatic brain injury and concussion: A concussion can disrupt fluid dynamics in the brain, but currently there is no direct way to measure that disruption. The AI velocimetry framework could offer a quantitative tool to assess whether glymphatic flow has been compromised after a head injury .
• Aging research: Glymphatic clearance naturally declines with age, contributing to cognitive decline . Directly comparing flow velocities in young versus old brains could uncover the physical mechanisms behind this decline and suggest interventions .

The method can also be adapted beyond imaging. The research group has already extended their modeling to study time-dependent flows for tracer injection and drug delivery simulations in the glymphatic network , suggesting future applications in guiding therapeutic delivery to the brain.

Physics-informed AI has given researchers their first real look at the brain’s waste-clearing plumbing in action. While applications in the clinic are still years away, the dual-speed flow map provides a quantitative foundation for understanding how the brain keeps itself clean—and what happens when that system fails.

The technical breakthrough: teaching AI the laws of fluid dynamics

Conventional MRI can show that a tracer dye is spreading through the brain, but it cannot measure the velocity of that spread when movement occurs at just a few microns per second . To overcome this, the researchers built neural networks and trained them on time-lapse MRI videos of dye dispersing through brain tissue. The key innovation was feeding the AI not just the imaging data, but also the fundamental equations of fluid dynamics .

This approach, known as physics-informed neural networks or artificial intelligence velocimetry (AIV), forces the model’s predictions to obey the laws of physics . By doing so, the AI could infer two previously inaccessible parameters from the dye’s movement: the local velocity of the fluid and the permeability of the surrounding brain tissue . The technique builds on earlier research by the same group that used AIV to quantify pressure, wall shear stress, and 3D velocities in murine perivascular spaces .

The discovery: a dual-speed clearance system

The AI-powered reconstruction exposed a stark contrast in how the glymphatic system moves fluid depending on location :

• Fast flow at the surface: In open regions, such as the subarachnoid space between the skull and the brain, CSF moves at a pace of a few microns per second, efficiently sweeping along major arterial highways.
• Slow flow deep in tissue: Once fluid penetrates the brain’s dense parenchyma (the functional tissue), it trickles through at a rate roughly 50 times slower than surface flow .

This dual-speed regime makes biological sense. The brain’s outer surface acts as a high-conductance distribution network, while deep tissue presents high hydraulic resistance, causing fluid to percolate slowly through narrow interstitial spaces . Previous modeling work by Kelley’s group had already pointed toward low-resistance perivascular spaces coupled with high-resistance parenchyma as the only configuration that could drive glymphatic flow with a small pressure drop and allow good perfusion throughout the cortex . The new AI measurements now provide direct in vivo evidence for this structure.

Beyond velocity: measuring tissue permeability

A major hidden variable in glymphatic research has been tissue permeability—how easily brain tissue allows fluid to pass through it. The new physics-informed AI framework simultaneously infers permeability by observing how the tracer disperses and constraining the solution with conservation laws . Changes in brain tissue permeability could be an early marker of pathology; if tissue becomes more resistant to fluid flow, waste clearance stalls. Being able to measure this property non-invasively from MRI could open a new window into the earliest stages of neurodegenerative disease.

Current limitations: the gap between mice and humans

It is critical to note that all measurements so far have been made in animal models, specifically mice, to establish baseline values . Human brain imaging presents significant additional hurdles, including larger scale, longer scan times, and the need for clinically safe tracer agents. The researchers are actively working toward adapting the method for human application, but this translational step remains a work in progress .

Clinical possibilities on the horizon

Even with these caveats, the long-term clinical possibilities are striking. The ability to directly measure glymphatic function from a standard MRI scan could one day transform neurology:

• Alzheimer’s disease screening: The glymphatic system is responsible for clearing amyloid beta proteins during sleep . If the AI method can be translated to humans, it could potentially measure whether an Alzheimer’s patient has impaired brain fluid circulation, or even screen healthy individuals earlier in life for poor clearance before plaques accumulate .
• Traumatic brain injury and concussion: A concussion can disrupt fluid dynamics in the brain, but currently there is no direct way to measure that disruption. The AI velocimetry framework could offer a quantitative tool to assess whether glymphatic flow has been compromised after a head injury .
• Aging research: Glymphatic clearance naturally declines with age, contributing to cognitive decline . Directly comparing flow velocities in young versus old brains could uncover the physical mechanisms behind this decline and suggest interventions .

The method can also be adapted beyond imaging. The research group has already extended their modeling to study time-dependent flows for tracer injection and drug delivery simulations in the glymphatic network , suggesting future applications in guiding therapeutic delivery to the brain.

Physics-informed AI has given researchers their first real look at the brain’s waste-clearing plumbing in action. While applications in the clinic are still years away, the dual-speed flow map provides a quantitative foundation for understanding how the brain keeps itself clean—and what happens when that system fails.