Composition of Classification Models for Recognizing the Flow Velocity of Liquids in Capillaries

Purpose of research. Development of a technique for estimating the average flow rate of physiological fluids in capillaries from images obtained using laser speckle-contrast imaging. The technique includes obtaining experimental data in the form of an image of the fluid flow in a thin tube, their preliminary processing, including filtering and compressing data, as well as training and testing approximate models using modern machine learning methods.Methods. The experimental study of the fluid flow in the tube is based on the application of the laser speckle-contrast imaging method. The spatial speckle-contrast values are calculated from the obtained images. The obtained data are subjected to preliminary processing, including the data filtering out and extending to a steady flow mode, as well as compressing the obtained images using the principal component method, which allows reducing the dimension of the feature space. The problem of predicting the average velocity from the image of the fluid flow is solved as a classification problem based on the composition of decision trees constructed through the bagging procedure, as well as in the form of a random forest.Results. A technique for predicting the average velocity of liquid flow in a capillary from images obtained using the laser speckle-contrast imaging method has been developed. The accuracy of predicting the average velocity (or flow rate) based on the training sample was about 91%, on the validation and test samples - at least 81.5%.Conclusion. Based on the developed technique, it is planned to determine the kinematic characteristics of the parameters of physiological fluids flow, which will improve the inertial method of measuring the viscosity of the tested liquids developed earlier by the authors, getting rid of a number of assumptions about the velocity profile.