The elasticity of red blood cells (RBCs) plays a vital role in their efficient movement through blood vessels, facilitating the transportation of oxygen within the bloodstream. However, various diseases significantly impact RBC elasticity, making it an important parameter for diagnosing and monitoring health conditions. In this study, we propose a novel approach to determine RBC elasticity by analyzing video recordings and using a convolutional neural network (CNN) for classification. Due to the scarcity of available blood flow recordings, computer simulations based on a numerical model are employed to generate a substantial amount of training data. The simulation model incorporates the representation of RBCs as elastic objects within a fluid flow, allowing for a detailed understanding of their behavior. We compare the performance of different CNN architectures, including ResNet and EfficientNet, for video classification of RBC elasticity. Our results demonstrate the potential of using CNNs and simulation-based data for the accurate classification of RBC elasticity.

RBC (Red Blood Cell) membrane is a highly elastic structure, and proper modelling of this elasticity is essential for biomedical applications that involve computational experiments with blood flow. In this work, we present a new method for estimating one of the key parameters of red blood cell elasticity, which uses a neural network trained on the simulation outputs. We test classic LSTM (Long-Short Term Memory) architecture for the time series regression task, and we also experiment with novel CNN-LSTM (Convolutional Neural Network) architecture. We paid special attention to investigating the impact of the way the three-dimensional training data are reduced to their two-dimensional projections. Such a comparison is possible thanks to working with simulation outputs that are equivalently defined for all dimensions and their combinations. The obtained results can be used as recommendations for an appropriate way to record real experiments for which the reduced dimension of the acquired data is essential.

The main aim of our work is future optimization of microfluidic devices used for capture of rare cells, for instance circulating tumour cells. Since the manufacture of such devices and performing biological experiments are complicated and time consuming, in silico experiments are preferable for this purpose. But first a simulation model has to be properly verified. In this article we propose statistical characteristics for comparison of simulation and biological experiments of red blood cells moving in a fluid. At the same time these characteristics allow better understanding of the course of the simulation and the behaviour of its individual components. In the first part we explore the sensitivity of some of the characteristics formulated on the grounds of the previous research to various topologies of the microfluidic channel with one solid hematocrit. The second part of the article analyzes the entire blood cell group; it focuses on their position (orientation) in the channel. To analyze the orientation, we use the principal component analysis (PCA), by means of which we can determine the statistical distribution of the directions of a blood cell set with a certain channel geometry. This characteristics describes the qualities of blood cells during a simulation, which are interesting when we are analyzing, as a targeted value, the possibility of the largest number possible of blood cells getting to the wall of the channel. We have tested the characteristics for various simulation parameters, and we have obtained interesting results.

The elasticity of red blood cells (RBCs) is crucial for their ability to fulfill their role in the blood. Decreased RBC deformability is associated with various pathological conditions. This study explores the application of machine learning to predict the elasticity of RBCs using both image data and detailed physical measurements derived from simulations. We simulated RBC behavior in a microfluidic channel. The simulation results provided the basis for generating data on which we applied machine learning techniques. We analyzed the surface-area-to-volume ratio of RBCs as an indicator of elasticity, employing statistical methods to differentiate between healthy and diseased RBCs. The Kolmogorov–Smirnov test confirmed significant differences between healthy and diseased RBCs, though distinctions among different types of diseased RBCs were less clear. We used decision tree models, including random forests and gradient boosting, to classify RBC elasticity based on predictors derived from simulation data. The comparison of the results with our previous work on deep neural networks shows improved classification accuracy in some scenarios. The study highlights the potential of machine learning to automate and enhance the analysis of RBC elasticity, with implications for clinical diagnostics.

In this work, we examine the volumetric flow rate of microfluidic devices. The volumetric flow rate is a parameter which is necessary to correctly set up a simulation of a real device and to check the conformity of a simulation and a laboratory experiments [1]. Instead of defining the volumetric rate at the beginning as a simulation parameter, a parameter of external force is set. The proposed hypothesis is that for a fixed set of other parameters (topology, viscosity of the liquid, …) the volumetric flow rate is linearly dependent on external force in typical ranges of fluid velocity used in our simulations. To confirm this linearity hypothesis and to find numerical limits of this approach, we test several values of the external force parameter. The tests are designed for three different topologies of simulation box and for various haematocrits. The topologies of the microfluidic devices are inspired by existing laboratory experiments [3 - 6]. The linear relationship between the external force and the volumetric flow rate is verified in orders of magnitudes similar to the values obtained from laboratory experiments.

Numerical models for the flow of blood and other fluids can be used to design and optimize microfluidic devices computationally and thus to save time and resources needed for production, testing, and redesigning of the physical microfluidic devices. Like biological experiments, computer simulations have their limitations. Data from both the biological and the computational experiments can be processed by machine learning methods to obtain new insights which then can be used for the optimization of the microfluidic devices and also for diagnostic purposes. In this work, we propose a method for identifying red blood cells in flow by their stiffness based on their movement data processed by neural networks. We describe the performed classification experiments and evaluate their accuracy in various modifications of the neural network model. We outline other uses of the model for processing data from video recordings of blood flow. The proposed model and neural network methodology classify healthy and more rigid (diseased) red blood cells with the accuracy of about 99.5% depending on the selected dataset that represents the flow of a suspension of blood cells of various levels of stiffness.