Artificial intelligence velocimetry for quantifying cerebral spinal fluid flow in the brain using physics-informed neural networks.

Penetrating perivascular spaces are key pathways that carry CSF deep into the brain tissue where solute exchange occurs, and failure of this process affects brain health and diseases such as Alzheimer's and small vessel disease. Measuring CSF flow in penetrating PVS is important for understanding brain waste clearance and nutrient distribution. Previously, researchers used particle tracking velocimetry (PTV) to measure speed and artificial intelligence velocimetry (AIV) to infer volume flow rates on the brain surface, but flow in penetrating PVSs has never been measured since the one-micron tracer particles used in PTV do not enter. In this study we demonstrated how AIV could be used to infer CSF flow in penetrating PVSs. We applied AIV to estimate flow rates in penetrating PVSs by inferring the flow at locations upstream and downstream of a bifurcation where a penetrating PVS diverges from a surface PVS. We first demonstrated the accuracy of the approach using synthetic data from a realistic PVS geometry where flow in the penetrating PVS is known. Then, using the same approach, we inferred the volume flow rate in penetrating PVSs for the first time.