A decision tree approach for classification of white blood cells based on image features

There are different types of blood cells in peripheral blood and body fluid. Each type of blood cells should maintain a certain proportion. When a person falls ill, some types of blood cells may present abnormal proportions due to different diseases. The most obvious change is in white blood cells (or WBC). Therefore, WBC differential count provides very important information for disease diagnosis. In this paper, an efficient decision tree is proposed for classifying white blood cells based on image features. White blood cells are segmented from a smear image and then classified into one of the six types based on the image features extracted from nucleus and cytoplasm. As compared to the inspection result obtained from a senior human specialist, the experimental result showed that 98.06% accuracy was achieved by using our method to inspect 1289 white blood cells contained in the smear images.