Application of machine learning methods to red blood cells motion detection

Abstract

The hypothesis of correlation between the blood cells motion in capillary and the body state is under study. Observation of the cells motion and deformation, their interaction between each other and with capillary walls may allow to extract an important information. The hypothesis corresponds to the results of some recent research that demonstrate correlation between cells viscoelastic properties and some of the diseases [1–5].
The present work deals with methods of the red blood cells (RBC) motion detection and velocity approximation. The videocapillaroscopy is applied as the experimental method to study blood microcirculation in a nailfold [6]. The methods of the research are based on both: the deterministic and the stochastic approaches. The deterministic approach of RBCs velocity approximation is based on the detection and the analysis of the intensity peaks on a sequence of images [6,7]. The stochastic approach is based on methods of machine learning, including semantic segmentation, object detection, feature extraction, and data fitting using deep convolutional networks, LSTM networks and some other network architectures [8].
The obtained results demonstrated positive effect of combination deterministic approach at the stage of data preprocessing and stochastic approach at the stages of feature extraction, object detection and velocity fitting.
This work was supported by the Russian Science Foundation under the Project 20-79-00332.


File with abstract

Speaker

Alexey Kornaev
Orel State University named after I.S. Turgenev
Russian Federation

Discussion

Ask question