CapillaryNet: An Automated System to Analyze Microcirculation Videos from Handheld Vital Microscopy

Capillaries are the smallest vessels in the body responsible for the delivery of oxygen and nutrients to the surrounding cells. Various diseases have been shown to alter the density of nutritive capillaries and the flow velocity of erythrocytes. In previous studies, capillary density and flow velocity have been assessed manually by trained specialists. Manual analysis of a 20-second long microvascular video takes on average 20 minutes and requires extensive training. Several studies have reported that manual analysis hinders the application of microvascular microscopy in a clinical setting. In this paper, we present a fully automated system, called CapillaryNet, that can automate microvascular microscopy analysis and thus enable the method to be used not just as a research tool, but also for clinical applications. Our method has been developed by acquiring microcirculation videos from 50 different subjects annotated by trained biomedical researchers. CapillaryNet detects capillaries with an accuracy comparable to trained researchers in less than 0.1% of the time taken by humans and measures several microvascular parameters that researchers were previously unable to quantify, i.e. capillary hematocrit and intra-capillary flow velocity heterogeneity.