Research In Implementation Of Automatic Blood Cell Classification System

With the development of medical science, people come to realize that cells in ourbody is the basic detection on our health. Therefore, cell-based image analysis methodsbecame important auxiliary tools for doctors to conduct clinical diagnosis, pathologicalanalysis and treatment.In traditional white blood cell tests, doctors must do microscopic observation,positioning and statistics artificially. The Hematology Analyzer currently used is basedon physical applications and made by foreign country. It also has versatility, samplepreservation, and expensive problems. It’s a very large workload, and very subjective. Inorder to improve efficiency, using image processing to do cell segmentation,classification and statistics is in need. Automatic blood cell classification system isdesigned to reach this goal.The paper describes the basic principles of blood medical testing firstly. Then, itintroduces the design and implementation methods of the blood cell classification system.Simultaneously, the paper puts an emphasis on WBC segmentation&classificationalgorithm and implementation. For WBC segmentation, I combine global threshold andadaptive threshold in local area, which makes segmentation more stable in different kindsof WBCs. For WBC classification, besides geometric characteristic values and colorcharacteristic values, I adds59LBP texture eigenvalues, which improves the effect ofWBC classification a lot. In the paper, I also do some deep research on how to choosegood training parameter values, and I give a detailed description on cross-validation andgrid-search.At the end of the paper, I introduce the significance of the system tests. At the sametime, I describe and analyze the effects of WBC segmentation and classification.