Research On Automatic White Blood Cell Recognition And Classification System

The immune system as the third line of defense of the human body protects the body from viruses, bacteria and so on, so it is crucial for the health of human body. The white blood cell which is a part of the immune system is responsible for the recognition and phagocytosis of abnormal cells, such as tumor cells. Normally, the content of each kind of white blood cells in the human body remains stable, and the imbalance of the content usually indicates healthy problems of human body. In the traditional blood routine examination, the white blood cells count based on classes and the morphological analysis mainly rely on manual work of experts, and it is a low efficiency and subjective job. Therefore, the development of the automatic classification and the morphological analysis blood cells is needed. The flow cytometer is one of the commonly used equipment to count the white blood cells. However, the flow cytometer is very expensive, and cannot recognize different kinds of white blood cells, so it has limitations in clinical application. In recent years, the classification and morphological analysis of white blood cells based on image processing and pattern recognition method have been developed gradually. The computer-aided classification and morphological analysis of white blood cells taking advantage of blood smear image is a technology of low cost, and can subdivide different white blood cell types, thus it has broad application prospects in clinic.At present, there are many methods to realize the computer-aided automatic white blood cell classification, but most of the methods cannot achieve satisfactory performance in practical application. Given that the computer-aided automatic white blood cell classification using blood smear image has wide application prospect in clinic, this paper proposes a complete white blood cell automatic classification system to achieve the subdivision of different white blood cell types and counting respectively, providing reliable basis for clinical diagnosis.The main content of this paper is to study and implement a complete white blood cell automatic classification system. In the researching process of the automatic classification system, the primary problem lies in the accurate segmentation of cytoplasm and nucleus of white blood cell, which is very difficult because of the non-uniform color and illumination of the blood smear image and the similarity between the color of the cytoplasm and the nucleus. In order to solve this problem, this paper proposes two solutions:one is a white blood cell segmentation algorithm based on color space transformation, and the other is a white blood cell segmentation algorithm using color-space-based K-means clustering. Each of them have its own advantages and disadvantages, and the details will be introduced in the main text. After completing the white blood cell segmentation, this paper adopts a new kind of feature extraction method to better represent various characteristics of the white blood cells. Then sequential forward selection algorithm is adopted to conduct the feature selection, removing redundant features that have little contribution to the classification, and it is a necessary step which helps to improve the classification accuracy and speed of classification. To realize the white blood cell classification, this paper adopts SVM classifier. White blood cell classification is a multi-classification problem, however, the SVM is essentially two-classes classifier, so this paper adopts the One-Against-All strategy to realize the multiple classification of white blood cells. Then the classification results are analyzed. Finally, this paper summarizes the research work, points out the existing problems and defects of this system, and discusses the plan for the future research.This paper proposes the complete system to classify, count and analyze the different kind of white blood cells in the blood smear image through image processing and pattern recognition method. The system includes the white blood cell segmentation, feature recognition, feature selection, and white blood cell classification. The method proposed in this paper, has better results compared with the former methods, and has high segmentation and classification accuracy and good robustness, and can adapt to different working environment. All these make it have a broad prospect of application in clinic.