Vision Transformer Network Based on Ultrasonic Radiofrequency Data for Red Blood Cell Aggregation Classification

The grading analysis of abnormal red blood cells (RBCs) aggregation holds significant clinical relevance, as RBC aggregation is closely associated with various pathological conditions such as hyperglycemia, acute myocardial infarction, coronary heart disease, and pulmonary embolism. The traditional method of measuring RBC aggregation by ultrasound is parameter estimation, but traditional parameter estimation methods require to neglect number of conditions and complicate calculations. To overcome the limitations of traditional methods, we make an appropriate adjustment to Vision Transformer (ViT) and apply it to the RBC aggregation classification. In this study, to develop the performance and versatility of the model, we prepared blood samples with varying degrees of RBC aggregation using 13 different plasma concentrations, with intervals of 5% between concentrations, and all acquired data are divided into training set, validation set, and test set as the ratio of 7:2:1. The results show that the model achieves 86.82% accuracy on the test set, and has the potential to provide a precise approach for non-invasive monitoring of RBC aggregation.