The Study Of Key Techniques Of Automated Blood Cell Morphological Analysis And Classification

Automated blood cell morphological analysis and recognition, has become an important means of clinical diagnosis, pathological analysis and treatment. It can help hematologist diagnose diseases, such as leukemia and blood cancer, and so on. The traditional method of white blood cell analysis and recognition is divided into three steps. Firstly, making the stain blood slide by medical experts; then, putting the stain blood smear under the oil microscope, finally, medical experts analysis and identify white blood cells under the microscope to diagnose whether the body has disease or not. The traditional microscopic examination method is of low efficiency, the technical requirements of medical experts are relatively high, and the recognition results also rely on expert's subjective judgment. There are a lot of limitations in clinical diagnosis of blood diseases. To develop an automated white blood cell morphological analysis and recognition system, using computer to analyse and classify white blood cells, has broad application prospects in clinical medicine.The morphological analysis and recognition researches for white blood cells are more, but there is no automatic blood cell morphology analysis and identification of instruments used in clinical trials depended on image processing and pattern recognition technology. There are a lot of drawbacks of the current existence of the white blood cell segmentation or recognition algorithms. Such as some algorithms cannot be a good solution to the complex problem of overlapping leukocyte, white blood cell segmentation and recognition accuracy is low and the algorithm robustness is not satisfactory, and so on. In view of the above problems, the paper presents an algorithm of automatic blood cell morphological analysis and recognition, which aims to develop an automatic blood cell recognition system.In this paper, the morphological analysis and classification of blood cell images, mainly including four aspects:sample database establishment, white blood cells segmentation, feature extraction, and white blood cells recognition.Due to physical, age, type of disease, light, dyeing conditions and different hardware equipment are hard to control, the quality of sample images are not the same. In the paper,24-bit RGB images were obtained for three types of smears (normal peripheral blood, M3 bone marrow, and M5 bone marrow) from more than 10 individuals.This paper adopts two methods to segment white blood cell. One is the method based on the color space and mathematical operations under a scope of illumination intensity, the other is based on mean shift clustering and adaptive threshold under different light intensity. The paper also adopts an improved watershed algorithm to segment complex overlapping leukocytes.In the paper, the characteristics of the white blood cells are characterized by the morphology, color and texture, in an amount small, reliable, independent and distinguishable feature extraction principle.White blood cell classification is a multi-classification problem. Through the analysis of a variety of classification algorithms, such as random forest, k-nearest neighbor, BP neural network, the paper finally decided to choose the classification strategy of the combination of random forest and K-nearest neighbor to identify the white blood cells. The recognition rate is better than other traditional algorithms. This paper develops a software system for automatic recognition of white blood cells to make the algorithm visualization under the MATLAB platform.This paper aims to propose a kind of automatic blood cell morphology analysis and classification algorithm. This paper established cell sample database; The results of experiments demonstrated that the algorithm can solve the problem of complex leukocyte adhesion; the algorithm is applicable to various cell sample database in segmentation and has a high segmentation rate; the selected features are well expressed in different types of white blood cells; the recognition strategies in this paper make the average recognition rate higher than other traditional recognition algorithms.