Automatic Flow Property Quantification on Time Series Images from PDMS Fabricated Veins

Accurate quantification of the fluid flow is crucial for studying the physiology of blood flow and associated pathologies like thrombosis. Here we generate 3D models of blood vessels in patients undergoing dialysis and fabricate fluidic devices (using PDMS) to perform flow experiments in vitro, using flow parameters measured in the same patients. We perform Particle Imaging Velocimetry on fluorescent tracer beads suspended in a fluid that matches the viscosity and density of blood. Videos of multiple regions of interest (ROI) of the fluidic model are used to characterize flow in vein regions that are likely to thrombose. Each bead moving along with the fluid displays a streamline highlighting the path along which it travels during a short exposure time. We obtain 25-30 images and hundreds of streamlines per ROI. Videos of multiple ROIs under different physiologic flow conditions are generated. We developed an automatic pipeline to extract the streamlines using computer vision algorithms. The streamlines are annotated with their length in pixels, angle orientation, and their distance from the vein wall from which we calculate wall shear stress (WSS), a critical predictor of thrombosis. For each ROI, we estimate WSS sampling rate for statistics. We apply the pipeline to 3D models of different geometry and flow conditions and obtain > 95% coverage on the vein wall boundary on average, for each ROI.