DEVELOPMENT AND RESEARCH OF COMPUTER VISION ALGORITHMS FOR IMAGE ANALYSIS OF DEFORMED ERYTHROCYTES

Abstract

The paper presents the results of the development and research of algorithms for analyzing images obtained by blood microscopy using computer vision technologies. The objects of the study are images of erythrocytes deformed in the shear flow. The deformability of erythrocytes largely determines the nature of blood microcirculation and therefore is directly related to the diagnosis and treatment of many diseases. At the current moment image analysis of the evaluation of blood cells deformations is usually performed visually by medical technicians. To automate the image processing process various technologies were investigated. As a result, a new computer vision approach for rapid and accurate recognition of erythrocytes deformed in shear flow is presented. The developed algorithms use image binarization and a neural network based on the U-Net architecture for separated erythrocytes, and a neural network based on the StarDist architecture for their conglomerates. We present the results of the algorithms on real lood microscopy images, compare their performance and discuss the practical applications in medical diagnostics. Evaluation of the distribution of erythrocytes by deformability will provide additional diagnostic and scientific information for medical research. Automation of image analysis of deformed images will increase the accuracy and simultaneously reduce the time and cost of tests, leading to a better patient treatment.