Leukemia Cell Feature Selection Methods For Research

Leukemia is a blood disease whose pathogenic mechanism is not yet clear, it does seriously harm to life and health of patients. With the continuous development and innovation in computer hardware and software technology, advanced image processing and pattern recognition technology has been widely used in medical institutions, and it provides a more scientific and standardized basis for clinical diagnosis. It is an important research topic in the field of medical image applications to identify the blood cell image using pattern recognition technology. Feature selection is crucial in the identification process of the blood cells. After preprocessing and segmentation, the features of leukemic cells are analyzed and selected in this paper. Those leukemia microscopic images are provided by a colleague of a medical college. The main work is as follows:First of all, on the basis of clinical identification to distinguish leukemic cells, we summarized the clinical characteristics of various types of leukemia cells. Distinguishable characteristics were put forward comprehensively.Secondly, features of leukemia cells were analyzed based on the distinguishable characteristics. The morphological features, color features and texture features of the leukemic cells were chosen comprehensively. There were totally35features, and the significance of those features was pointed out. Then the BP neural network was used to verify the effectiveness of chosen features, and better recognition result was achieved..Finally, since the number of features of the leukemia cell is very large, the feature dimensionality should be reduced according to their contributions. An improved ReliefF algorithm was proposed which can cover all the original sample set. Using the furthest samples instead of the closest neighbor samples in the same sample class, the distinguishable features'weights can be accumulated. However, the redundancy can not be removed with the ReliefF algorithm, and Mitra algorithm can be used to remove the redundancy effectively. A combination of feature selection algorithm based on the improved ReliefF algorithm and Mitra algorithm was proposed. We used the improved reliefF algorithm to select some most important features firstly. And then, Mitra algorithm was used to remove the redundancy. After that, an ideal feature set came out. BP neural network was also used to verify the effectiveness of the above methods.