Influence of CNN Input Modification for Red Blood Cells Trajectory Prediction in Blood Flow

Computer simulations of blood flow in microfluidic devices help for their development. These simulations are limited by their computational complexity. A possible solution to this problem is to analyze the simulation output data using machine learning methods. We use convolutional neural networks (CNN) for red blood cells (RBCs) trajectory prediction, which is very important for blood flow modeling. In this paper, we study how miscellaneous modifications of a CNN input affect the results from the learning experiments. All performed neural network experiments have sufficient accuracy. We evaluated, which CNN input parameters and their values have a significant impact on experiments accuracy.