Deep Learning-Based Assessment of Non-Physiological Damage in Red Blood Cells Through Morphological Characterization
Youyuan Xu, Yichong Hu, Yue Yu, Ying Li, Lining Sun, Hao Chun Yang · 2025
Cardiovascular disease in now one of the leading causes of death worldwide. And artificial heart has become one of the effective methods for the treatment of cardiovascular diseases. In artificial hearts, it is essential to study the pattern of mechanical damage to the blood, and how to quickly and accurately determine the degree of mechanical damage to the blood has become a crucial problem. Here, we propose a neural network-based approach to study the relationship between mechanical damage and morphological characteristics experienced by red blood cells (RBCs). We used the Taylor-Couette shear device to make mechanical damage to the blood. We use a dielectrophoretic microfluidic method to capture RBCs and photograph microscopic images. A convolutional neural network model was trained using the obtained image dataset for final cell classification. The proposed model achieves an accuracy rate of more than 90%.