Red Blood Cell Antigen Typing Based on Image Processing and Machine Learning

Red blood cell (RBC) antigen typing is mainly used in blood group testing, which is critical to providing compatible blood and minimizing the risk of hemolytic transfusion reactions in blood transfusion. This study developed a procedure based on image processing and machine learning, to automatically determinate red blood cell antigen type from the RBC agglutination images. Red blood cell agglutination samples were prepared by mixing blood samples with antibodies in a multi-channel microfluidic device and then captured with a high-speed document scanner. Each image contains 24 blood agglutination sub-areas, which would be automatically cut into 24 sub-images and analyzed in the procedure. HOG (Histogram of oriented gradients), Grayscale Histogram and LBP (Local Binary Pattern) features have been extracted from the sub-images and combined with machine learning algorithms for classification. The machine learning algorithms used in this study included k-Nearest Neighbors (K-NN), Support Vector Machine (SVM), Logistic Regression and Multilayer Perceptron (MLP). Among them, the average accuracy rate of the combination of SVM and histogram reached 93.41% in determining the blood agglutination patterns (agglutination or no agglutination, binary classification), and the accuracy has reached more than 98% on some data sets. Besides, the study adopted a hierarchical classification strategy using different features and achieved an accuracy of 86.57% in detecting blood agglutination degrees (5 classification).