Automated Red Blood Cell Classification with Deep Learning
Inkyu Moon · 2022
Red Blood Cell (RBC) transfusions are a life-saving clinical procedure for patients with severe bleeding before or during surgery. This chapter introduces new deep learning-based methods for efficient RBC segmentation and classification to reduce the computational burden while achieving a high classification accuracy. It shows that the presented deep-learning models can classify RBCs stored for different durations, identify dominant shapes in each storage group, and evaluate storage lesions in RBCs for safe transfusions. Convolutional neural networks are widely used in computer vision and visual recognition problems with remarkable results. These methods have been applied to RBC segmentation with good performance outcomes. Some challenges in RBC segmentation include independent boundary detection and the separation of overlapping RBCs. Increased storage duration was strongly related to the transformation of RBCs from discocytes to transitory echinocytes and finally to spherocytes.